test(conformance): pin section 11's anchors and transclusions

[CameronBrooks11]

Adds 215 conformance cases of the anchor kind under conformance/cases/anchor/, pinning what SPEC.md §11 decides about anchors and transclusion. It also adds their provenance entry to conformance/cases/README.md.

How they were written

The cases come from SPEC.md (§11 above all), docs/adjudications/0003-anchors.md with all three of its addenda, 0001 and 0002, the case-tree README, case-body.schema.json and the tree on main at 66ce7f1, and later at ed2e50e (#47) and f75d069 (#48), the second and third addenda, which this branch has merged in. I did not read any implementation of §11: not #45, not its branch, not any branch but main. The only code I read was the runner, check_tree.py, anchor_cases.py and the block reader reference/blockspec/document.py, to learn the parse.

Each WHY.md quotes the passage that decides its outcome. I cross-checked every status, leaf and span with a throwaway calculation of §11.2 to §11.5, written from the spec and not committed. It ran on the repository that anchor_cases.Repository builds, so placeholders were hashed by git and merges were run by stock git, as the runner does them. Every merge produced the outcome its case states.

Landing plan

main's reference has no resolve, so all 215 anchor cases fail just conform with "the implementation defines no resolve(repo, queries)", and just test fails on the same root cause (test_conformance, plus the mutation probe's baseline). This is the order the README's Landing order decides. The cases are reviewed here. Then #45 rebases onto them and runs against them with no case changed, and that merge lands the cases and the resolver together. This PR then closes as landed through #45. Do not merge this PR on its own.

run_cases.MIN_CASES is still 237. I did not change it, because this PR touches only conformance/cases/. Raise it to 452 when the cases land.

just check, just tree (452 cases, 0 problems), kindkit's validate_case_tree.py and schema_selftest.py all pass.

What the cases cover

record/refuses-an-unknown-leaf-type cannot fail for the T check alone: the leaf-at-nth rule refuses the same record, and its WHY.md says so.

group	cases	what
status/	12	RESOLVED, MOVED, EDITED, DETACHED, ORPHANED; a heading that only changes level; CRLF and NFD re-saves
leaf/	5	leaves in a list item, a block quote, a code block and a table; the | escape keeping a one-column and a two-column table apart
context/	6	a short prefix that must sit at the leaf's start; a short suffix that may, and must, end at the leaf's end; an edit one code point outside and on the first code point of a full prefix; code-point offsets with a character outside the BMP
record/	39	the canonical form accepted (every escape, U+0080 escaped, SHA-256) and refused: an anchor that is a JSON array, a quote that overlaps itself, U+0080 written raw, a negative nth, a context of 33 code points, a body the profile refuses, notes file names that are not <ULID>.md (not a ULID, lower case, a letter outside Crockford's set, a first character above 7, .txt), no frontmatter, added or reordered lines, unknown status, trailing space, spacing, key order, other escapes, upper-case hex escape, raw U+007F, \t, unquoted or duplicate key, fractional nth, unknown T, replace on a comment or missing on a suggestion, short contexts away from the edge, empty span, a triple that does not occur or occurs twice, nth past the copies
input/	7	not-in-profile for a base with a setext heading; document-missing (deleted; renamed without its notes), base-unavailable (after a squash; an id no file has), not-in-profile (raw HTML; committed conflict markers)
correspondence/	7	the leftmost chain on a two-leaf swap; the 0.15 margin on the base leaf's side; the gap alignment skipping the base element first; similarity exactly 0.5 and just below; a margin of exactly 0.15 (where IEEE 754 doubles get it wrong) and just below
no-pair/, repeat/, near-copy/, refilled-gap/, exact-copy/, crowded-gap/, span/	18	each §11.4 and §11.5 rule with its controls: twins kept apart by unique neighbours, twins beside a moved leaf, a twin deleted, a copy added, near-copies on EXACT (including a similarity of exactly 0.5, and a verbatim copy, similarity 1, that is not one), MOVED and EDITED, the refill rule, the Q12 narrowing, a crowded gap, two hits, two overlapping hits, and an alignment that detaches
suggestion/	35	unmappable-offset for a span across or ending right before the spaces a partly consumed tab leaves in a code block (in a list item, and under a fence's own indentation) and for a span ending in a code block's final \n, with a span beginning after the leftover spaces and one ending before the final \n that apply; write-back in a leaf with a tab of list-item indentation (a span right after it, mid-line after it, and before it), and after a marker followed by a tab; unmappable-offset on a list item's lazy line and under a quote's > and tab; unmappable-offset for a \n in replace on a list item's first line, a lazy line, a tab-prefixed line and a last line with no ending (with its control); the gate (RESOLVED, EDITED and heading level apply; MOVED, DETACHED, ORPHANED, refused, accepted, rejected and resolved are stale) and exact write-back: CRLF, a line break as CRLF and as a bare CR, a span across a CRLF break, > and list-item continuation prefixes, no normalisation of the leaf or of replace, and unmappable-offset
reply/	15	replies whose document is missing or refused, a malformed reply-to before document-missing, and document-missing before a missing parent; reply-to lines with extra spacing or single quotes refused; a lower-case and a non-ULID reply-to refused; replies to a record, to a reply, to an orphaned record; #REF!(reply-to) for a missing parent and for a parent in another notes directory; a bare id and a status line refused
transclusion/	30	under CommonMark 0.31.2, no transclusion for a link inside link text, a title &#116ransclude (no ;, so not an entity), an image, or prose holding "transclude"), and a transclusion whose title is followed by a line ending before ); a marker name that is also a heading's slug resolving to the marker; a heading slug that matches while no marker does (#REF!(name)); resolution after edits, of a region, after a move with its name; name-rebound by insertion and by a moved marker, and re-pinning; pinned-orphaned, pinned-crowded-gap, pinned-near-copy; #REF!(name); a pinned name that binds nothing; unresolvable-path (/, a scheme, .. above the root); .. and . from a subdirectory; document-missing (absent, case, percent-encoding); base-unavailable; not-in-profile for the target and the pinned blob; a link with another title
pins/	10	two sorted pins; pins matched by raw path (./policy.md and policy.md with different outcomes); a list item after another key, not refused; then each grammar break refusing every transclusion: order, duplicate, spacing, a pin after a comment line or a blank line, two pins: lines, a pins: line with more after it
merge/	7	clean merges whose result resolves (other leaves; two lines of the anchored leaf, with a suggestion that applies; a status change beside a re-anchoring; re-pins of two targets); conflicts giving not-in-profile (a document and a transclusion target) and record-inconsistent (two re-anchorings)
order/	7	§11.10's record order, each case setting up two reasons: grammar and a bad file name before document-missing; document-missing and a refused current text before base-unavailable; base-unavailable before the leaf-at-nth check; a quote that occurs twice and a short context (step 1) before base-unavailable
transclusion-order/	10	an unbound name (no marker, no slug) after base-unavailable; a pinned-<reason> before name-rebound; §11.9's step order: unresolvable-path before broken pins; broken pins before document-missing; document-missing and a refused target before base-unavailable; a slug-only name before base-unavailable and before the pinned binding; a refused pinned blob before the pinned binding; several pinned-<reason>s reporting the first leaf
referrer/	3	the referrer-level REFUSED(not-in-profile): raw HTML, a conflict line in a fence, and two adjacent re-pins that conflict under stock git
residual/	4	R1 to R4, pinned as §11.12 says they resolve. Each WHY.md marks the outcome as a documented residual, not a defect

Questions §11 did not decide

The first version of this PR listed nine. The second addendum (#47) decides all of them, and the cases above pin each decision: a context over 32, the order of input reasons (records and transclusions), unresolvable-path before broken pins, the line breaks that cannot be written back, notes files and reply-to ids that are not ULIDs, a slug-only name, a refused or conflicted referrer, a refused record body, and the unreachable #REF!(name) and EXACT gap clauses, which are now removed. I rechecked every existing case against the new text. No expected outcome changed. Twenty-six WHY.md files quoted wording that the addendum changed, and they now quote the current text.

The two questions this PR later raised, overlapping occurrences and a span across a tab that list indentation partly consumes, are decided by the third addendum (#48), and cases now pin both. Under it I rechecked suggestion/writes-back-after-a-tab-in-an-item-s-indentation: in a paragraph the leftover tab columns are indentation the line loses, so they never enter leaf.text, and the span maps as before.

The open items §12 still lists have no case here: whether creation and the consistency check count under the edge rule, further frontmatter keys, a pin no transclusion uses, frontmatter that hides the pins line, write-back into a table, renderer-style path decoding and empty segments, symbolic links and submodules, and the transclusion details (no path, angle brackets, other title quoting, a link among other text, a transclusion inside a list item or a quote, no pin, self-transclusion). A subfolder in a notes directory is decided but can't be expressed: the case kind names records by file, so no case pins it.

[CameronBrooks11]

Landed through #45, which merged this branch's final head (d389dee) together with the resolver, as the landing order in conformance/cases/README.md describes.