feat(testing): execute and amplify local tests with explicit query bounds #1797

Bounded property-testing milestone from ROADMAP's communication/query/learning/testing area; also supplies the BLR non-parity client requested in #614 G2. Worker codex-roadmap-learning-20261030 owns this task on branch feat/testing-local-repeat-20261030 in the provisioned isolated Extra-drive learning lane.

Reuse the existing PCP.LocalTest access model, its BLR instance PCP.HadamardTester.blrTest, DecisionTree, and BitString.hammingDist/BLR soundness. Do not introduce a parallel tester, probability, or distance model. Add a small reusable execution bridge from Boolean local tests to decision trees with depth at most the query-layout cardinality, and independent repetition accepting iff all trials accept, with exact acceptance probability a^k and at most k * |Q| queries. Apply this to the existing BLR instance: parities always pass, while a function at relative Hamming distance greater than ε from every parity passes k independent tests with probability at most (1-ε)^k (0 ≤ ε ≤ 1). The explicit far-from-property promise is essential; this is not exact recognition of arbitrary linearity with bounded error independent of ε.

Acceptance: documented reusable API in small topic modules; checked BLR tree runs and a concrete non-parity with nontrivial acceptance/distance; zero repetitions and empty-coordinate edge cases; full six default gates, hygiene/import checks, self-review, and outside-author-fleet review of the exact final revision. Existing definitions considered are listed above; no new general model is needed. No headline proposal: assess curated abstracts as unchanged and preserve the broader open ROADMAP work on tolerant/distribution testers, access models, lower bounds and native learner running times.

Mathematical core reuses the existing proved BLR theorem; new execution/repetition lemmas are original routine constructions. AI assistance: OpenAI Codex implements and validates the change under the user's parallel-roadmap request.