docs(adjudication): decide 0003's input-problem order and gaps

[CameronBrooks11]

Adds a second addendum to docs/adjudications/0003-anchors.md, recording
the owner's decisions of 2026-10-16 on nine gaps §11 left open, and
carries each into SPEC.md §11. No implementation and no conformance case
changes; the anchor case kind gains one shape (below).

The owner's decisions

	decided	SPEC
1	a prefix or suffix longer than 32 code points is record-inconsistent	§11.1
2	several input problems report the first in data-dependency order: record grammar, document-missing, document not-in-profile, base-unavailable, base not-in-profile, leaf at nth (record-inconsistent); several pinned-<reason>s the same way (the leaves taken in base order is this write-up's precision, below)	§11.10
3	unresolvable-path before a broken pins block	§11.9
4	a \n in replace that cannot be re-emitted (a list item's first line, a lazy line, a tab in the container prefix, the last line of a file with no final line ending) is stale unmappable-offset	§11.7
5	a notes-directory file not named <ULID>.md, a subfolder, or a malformed reply-to is record-inconsistent	§11.1
6	a name that only a heading slug matches in the current target is #REF!(name), checked before the pinned blob's binding; "binds" means marker binding throughout §11.9	§11.9
7	a referrer with conflict markers refuses its transclusions as not-in-profile; a referrer the profile refuses is refused as a whole and its transclusions are not enumerated; a conflict-marker line anywhere counts, frontmatter and fences included, whatever readings 4 and 12 say; §11.10's row names the referrer	§11.9, §11.10
8	a record whose body the profile refuses is record-inconsistent	§11.1
9	two clauses that can never apply are removed (proofs below)	§11.4, §11.9, §11.10

Precision the write-up adds, marked as not the owner's

Each is marked in the addendum with its reason.

• Which checks go in item 2's two record-inconsistent steps. Step 1
  covers everything that reads only the record, leaf.text included: file
  name, frontmatter, body, an empty span, an over-long context, the
  exact-once occurrence of prefix + span + suffix, and the short-context
  edge check. Step 6 is only the leaf at nth in base. So a triple that
  occurs twice in leaf.text, with an unreadable base, is
  record-inconsistent. Reason: each check sits at the first step whose
  data it reads.
• The full order for a transclusion, now §11.9's numbered steps:
  referrer not-in-profile, unresolvable-path, broken pins,
  document-missing, current-target not-in-profile, slug-only name
  #REF!(name), pinned blob base-unavailable, pinned blob
  not-in-profile, name unbound in the pinned blob, then the name, leaf and
  rebinding steps. Reason: the path and the pins are read from the
  referrer, and the rest follows the record's order with the pin standing
  in for base. This supersedes the first addendum's detail 7, which put
  the path before everything else.
• Pinned leaves in base order. Among several pinned-<reason>s, the
  first leaf's in the pinned blob's document order is reported. Reason:
  that is the order the leaf step visits them.
• The ULID form. 26 upper-case Crockford base-32 characters, the first
  0 to 7. Reason: one spelling per id, so ids, file names and
  reply-to compare byte for byte.

This closes the §12 item about conflicts among pin lines, which is removed.

The two proofs (item 9)

• #REF!(name) for "more than one" match. §11.9 requires the current
  target to pass the profile before the name resolves, the profile refuses
  a duplicate name, and the name is matched against markers only.
• "The pair must lie in the gap" under EXACT. The rule fires only
  for a repeated key. A repeated key is never on the chain, so the pair
  comes from step 2's alignment inside the anchored leaf's own gap.

0003's body keeps both clauses as first written; the addendum removes them.

Checks

$ just check
All checks passed!
$ just test
27 passed
$ python3 conformance/schema_selftest.py <kindkit at 6ebd80d>
0 broken body(ies) not rejected
$ just conform
0 failure(s) across 237 case(s) in the fixture tree
$ just tree
0 problem(s)

The anchor case kind

• conformance/cases/README.md: the ordering note, now Cases that set up
  several reasons, lets a case set up several input reasons and expect the
  first by §11.10's order, and keeps a single-reason case as the clearest
  way to pin one reason. A referrer refused as a whole is written as the
  one string "REFUSED(not-in-profile)" instead of a list. The limits list
  now says a subfolder in a notes directory cannot be named.
• case-body.schema.json accepts that string as a referrer's entry.
  schema_selftest.py's good body has both shapes, and two new broken
  bodies are rejected: another referrer-level status, and the status inside
  a list. Run against kindkit at the pinned 6ebd80d, 0 broken bodies are
  accepted.
• anchor_cases.py compares the string, builds alterations for it, and
  counts it as a refusal when deciding whether a case needs a control.
  tests/test_anchor_kind.py covers this, and the test fails when the
  comparison is disabled.

Not merged: this awaits review.