fix(memory): restore executable graph ingest policy

[dmitrii-f-t27]

The merged memory graph ingest source fails to parse and its test_* functions are not executable tests. Restore slice 4 as a pure extraction policy that calls the actual scope, temporal and fact-validation signatures and returns one atomic outcome.

The policy rechecks admission at emission and commit, requires explicit domain consent before disclosure, retries through RECEIVED with a three-request limit, and makes both terminal states absorbing. Twelve executable tests cover invalid codes, 64 state/event pairs, 35 domain/consent pairs, revoked consent and complete retry exhaustion.

Closes #8752. References #8425 and #7828.

Validation: t27c/Zig and native arm64 t27b with JIT/interpreter comparison both pass 12/12 tests and 574 runtime assertions. A wrong expected outcome and a widened transition each fail exactly one test, exit 1. The generated seal verifies the spec, closure and all four backend hashes. No foreign implementation or CI gate change.

The effect is a policy request only: no LLM, production data, KMS, service, FPGA or reward claim. The caller is responsible for supplying a fresh admission verdict. Full CI and ordinary merge remain pending. Evidence and all task-chain dispositions: docs/reports/2026-10-10-memory-graph-ingest-8752.md.

[dmitrii-f-t27]

@gHashTag this PR is ready for merge from my side. Head 9f4645a5.

What changed since the last review:

• a8e0e488 merged master and resolved the conflict with corpus: repair the five unexpected failures at 760a787 + dupe-bodies ledger + reseals (Closes #8957) #8962 in favour of the executable ingest policy. Checked with t27c built from this tree: test-report 12/12 pass, seal --verify all MATCH.
• 9f4645a5 the PR removes the duplicate struct Episode from specs/memory/graph/ingest.t27, so Episode is no longer a conflicted type name. Removed it from docs/reports/type_conflicts.json (tri types ratchet --bless), from type_conflicts_classified.json and from the table in docs/TYPE_CONFLICTS.md (47 DISTINCT). Run locally: tri types ratchet CLEAN, tri types classified OK, t27c suite --ratchet --corpus-only CLEAN.

CI on 9f4645a5: all four required checks pass, and Corpus ratchet passes. Two advisory checks are red. They also fail on master, so this PR did not cause them:

• cli-tri: cli/tri/src/gates.rs:9012 does include_str! on .github/workflows/brain-seal-refresh.yml, and that file is no longer on master.
• duplicate-bodies: the same FPGA body groups (bit_at, build_level, led_state_next, …) fail on master 010e26c0.

I have no merge rights here. Could you merge it when you have a moment?

Dmitrii Fedorov (@dmitrii-f-t27)