fix(release): prepare 4.10.12 memory reliability candidate

[kevintseng]

Summary

Prepare 4.10.12 for the next release channel with reliable backup/restore, task-state freshness, failure reporting, and lesson recall. Stop capture requires a matching successful edit result and records neutral observations instead of claiming that a failed or unanswered edit fixed an error. Align release verification with the Dashboard UI waiver while retaining shared API/data tests, source/security checks, coverage floors, package installation and upgrade checks.

Type of change

• Bug fix (fix)
• Build / CI / chore
• Release (release)

Docs synced

• CHANGELOG.md updated
• Architecture and API documentation updated
• README and README.zh-TW.md synchronized
• Version anchors consistently declare 4.10.12
• dist/skills-manifest.json regenerated and installed runtime hashes checked
• Doctor overall clean: controlled installed profiles retain diagnostic warnings, reported as PASS_WITH_CONCERNS

Verification

Final local candidate: ee51bce9827185a620e0217a003dece9796e5bee, tree 8f76a5e35f5b7b6464d817c677bb949dc9b89bea.

npm run verify exited 0: build, release gates, isolated non-UI tests, and packaged artifact smoke all passed. 280 test files passed; 5607 tests passed; 9 skipped. The tree did not change during verification.

[verify] tree 8f76a5e35f5b7b6464d817c677bb949dc9b89bea (full)
[verify] GREEN. Receipt for tree 8f76a5e35f5b written to .verify/receipt.json.

• Typecheck, lint, build, version coherence and source/security gates passed
• Real packed installation, CLI/doctor, MCP protocol and controlled routing smoke passed
• Stop failed/missing/success result regression evidence and installed-hook replay retained
• Affected capture fixtures: 33 passed; release workflow contracts: 54 passed; CLI fixture compatibility: 30 passed
• Independent affected-delta review completed; original failures retained
• CI fixture compatibility: isolate the error-envelope cases from the independently tested update worker, retain Windows unsupported-host refusal assertions, and use a file URL for the ESM refresh fixture. No product behavior or safety gate changed.

Test plan

• Run npm run verify on the clean exact candidate and require its real exit code and matching tree receipt
• Verify installed runtime manifest entries and preserve applicable unchanged host/restore evidence
• Wait for hosted CI, including the unchanged coverage floors
• After normal merge, run the supported release preconditions for --prerelease; publish to next only

Known limitations / follow-ups

Dashboard UI functional verification is explicitly SKIP, not PASS. The nine Windows-specific skipped cases remain unverified locally. Existing native-host diagnostic concerns and prior failed-run history are retained; unchanged evidence is reused with explicit applicability rather than represented as a fresh model journey. BEL/RLO display robustness remains deferred, not fixed.

HTTP forget now returns 404 for a missing memory and 400 for a non-matching observation. Clients relying on the old successful no-op response must handle these failures. latest promotion is separate and is not part of this PR.

Coverage

Surface	QA	Review	Simplification
Memory persistence, backup/restore, task-state freshness and failure semantics	Applicable prior effects plus exact-tree full verification	Existing whole-candidate and affected-delta review retained	Bounded fixes; broad tool/architecture changes deferred
Hook capture and lesson recall	Failed/missing/success regression cases, installed replay, exact-tree tests	Independent hook/delta review	Matching tool results; no new framework
Release entry points, CI and lifecycle	172 release-entry tests, 54 workflow contracts, final full verify	Independent deltas; warnings retained	Shared existing Vitest configuration; original gates retained
Generated runtime assets and documentation	Build, coherence/parity/documentation gates, installed manifest integrity	Source/generated consistency reviewed	Generated assets; no new runtime mechanism