fix(specs): restore executable policies and corpus checks

[dmitrii-f-t27]

Several existing specs block native test execution and the corpus, duplicate-body and generation checks. This change restores executable viewport/counter tests and the memory graph extraction policy, removes parser-discarded optimizer placeholders, and reuses existing AST/string helpers instead of duplicate function bodies.

Closes #8745. Closes #8752. Closes #8756. Closes #8767. References #7641, #8425 and #7828.

• Viewport arithmetic preserves the public byte-sized schema while avoiding overflow. The counter/LED model uses Boolean comparison and the actual 23-bit wrap, with boundary regressions.
• Memory graph ingest calls the real scope, retention and fact-validation signatures. One atomic outcome controls effect emission, commit and refusal; disclosure requires explicit domain consent, retries require fresh admission, and terminal states absorb repeated events.
• Two seal fixtures parse completely and use distinct module names for separate seals. The optimizer keeps its two deferred requirements as prose instead of fake nonexecuting tests. AST and plan specs reuse existing helpers. The synthetic lookup fixture observes the original ten command lengths through a numeric table that generates Verilog.

The structure validator uses native T27 byte scans while preserving exact keys, trimming, comments and arbitrary values; its phi fixture uses a bounded numerical approximation with a residual check. Seven boundary vectors cover empty input, CRLF/tabs, malformed delimiters, key case/prefixes and duplicates.

Validation: 6+7 viewport/counter tests, 12 graph tests, 58 parser/helper/loop tests and 15 structure-validator tests pass. The current website viewport consumer additionally checks all 43 assertions and its unchanged schema. Graph and the seven executable corpus-repair specs pass native arm64 t27b with JIT/interpreter comparison. Full cargo test --release -p t27b passes after synchronizing with upstream cd569e3. Generated seals verify all four backend hashes. Duplicate-body ratchet passes without ledger changes. Wrong expected outcomes, counter mask removal, widened emission and a wrong command length produce actual test failures.

The measured 1824-spec reference corpus before the upstream refresh is CLEAN against its existing ledger: 71 known primary failures match its ledger, with zero unexpected failures/passes or gate drift. This is not an all-passing corpus claim. L2 regeneration verifies all six relevant tracked outputs byte for byte. Seal coverage/currency and both type-conflict checks pass locally. The current upstream native cap 10 is retained; four newly measured pass records are added. Latest-head remote CI remains pending after this final source repair. No gate has been weakened or wholesale baseline blessed. Seven other specs in #7641 remain blocked, so that parent stays open. The optimizer's native implementation and the compiler's general seal-name collision bug are not claimed as fixed. No physical board, complete inference, real LLM call, production KMS, reward, real command-lookup performance or service deployment is claimed.

Reports: docs/reports/2026-10-10-spec-test-restoration-8745.md, docs/reports/2026-10-10-memory-graph-ingest-8752.md, docs/reports/2026-10-10-spec-corpus-repair-8756.md, docs/reports/2026-10-10-spec-structure-native-8767.md. All repair source commits are by Dmitrii Fedorov (dmitrii-f-t27); genuine upstream authorship is preserved.

[dmitrii-f-t27]

Status at head c2bd66d6d0cc6292c762287ceaf3192dd90b1f61 (updated 2026-10-11; replaces the earlier Verilog-widths note, which no longer applies).

What changed since 9e6cb1b98:

• 7b520bbec merges upstream master 5e5493cab (no rebase, no force-push). The only conflicts were three seals; ast_scan, ast_shape and bogatyrs_spec_structure are resealed with t27c built from the merged tree (Zig 0.16; tests 11/11, 13/13, 15/15). The other 91 seals touched by either side verify all hashes MATCH.
• c2bd66d6d regenerates gen/rust/tri/t27b/{ast_scan,ast_shape,coerce_plan,const_div_plan}.rs with that t27c: upstream changed how gen-rust emits slice parameters, and L2 GENERATION flagged the old copies. tools/l2_regen_check.py: all 6 checked files are t27c output; cargo test --release -p t27b passes.

Required checks at this head: validate, check-linked-issue, parse-ratchet, own-language are green, as are Check L1 TRACEABILITY and emit-bitexact (26 success / 5 failure / 0 pending). Mergeable.

The five red checks fail the same way on upstream master 5e5493cab:

Check	This PR	Master
Corpus ratchet (types)	Entry, Inputs, SkillMeta NEW conflict (job)	same three names (run 38146939229)
spec-guards	1 stale seal: queen_QueenHolds.json gen_hash_rust 7a85ae31df5c vs a028b1872114 (job)	same seal, same hashes (run 38146939153)
coverage	1 seal newly does not hold, same QueenHolds drift (job)	same (run 38146939200)
t27b-native-ratchet	UNLISTED 17, 0 unexpected failures or passes (job)	UNLISTED 17, the same 17 paths (run 38146939191)
cli-tri	census quiet 185 -> 188, shell 305 -> 306 (job)	same numbers on master ancestor a31427373 (run 38145387223); tri census pin --gate on a 5e5493cab checkout prints the same two moves

This PR changes no file under .github/ and does not touch specs/queen/holds.t27. t27c seal specs/queen/holds.t27 --verify with t27c built from master 5e5493cab reports the same gen_hash_rust MISMATCH. I did not reseal it here, because that would change a spec outside this PR's scope.

Dependent PR #8771 now contains this head (merge 35496172f) and shows the same five red checks.

Dmitrii Fedorov (@dmitrii-f-t27)