feat(history): preserve ordinary durable updates in immutable temporal history

[DivyamTalwar]

Original ZIP gap

Implements C03 / GAP-03, refs #52. Stacked on #58 for a reproducible reviewed sequence; no dependency on another user product. The earlier manual J09 perspective store is retained, but ordinary durable writes now have their own supported historical path.

Implementation

• Record fact/page history, immutable effect identities, operation receipt and projection event in the same SQLite transaction as each ordinary durable add/update/delete/noop.
• Add bounded snapshot_at() reads by transaction revision/time and valid time, immutable historical-page lookup, and an explicit structured correction path through commit_temporal().
• Preserve earlier knowledge perspectives when current validity changes. Exact retries do not add a second historical effect.
• Add explicit v1-to-v2 history enablement with a known coverage baseline; unavailable pre-migration history is not invented.
• Expose bounded outbox_events() and get_event() for independently delivered consumers, regardless of the reference consumer's delivered bit.
• Provide compact-copy history retirement into a new image while preserving original stores, receipts and retained source pages.

Review-driven corrections

An independent frozen-source review found import/migration provenance gaps, a compatibility regression, and incomplete corruption detection. The final candidate reconciles fact/page scope, preserves split provenance, validates immutable effects and indexed payload identity, rejects overlapping migration baselines atomically, and bounds event/page acquisition before loading payloads. Current corrected validity is distinct from independent page expiry/TTL.

Applicable independent probes pass; two older probe expectations are explicitly superseded by stronger migration rejection/unknown-coverage contracts, not silently deleted. Exact findings, replacement controls and limitations are in docs/closure/C03-remediation.md.

Actual assembled verification

Candidate: 5dc8e4855d74042964fdbfa8074350d4ea372af3.
Parent: 3803a448bb4dfa493a2d8e43999f73cf9a23443c.
Supervisor rerun on unchanged source bytes:

PYTHONPATH=$PWD python -m pytest -q -p no:cacheprovider
# 1016 passed, 0 failed, 0 skipped; 3 expected optional-provider warnings
python -m compileall -q jitmind tests
python -m pip check
python -m ruff check --no-cache --isolated --select E9,F <owned Python files>
git diff --check
# all passed

Tests use actual SQLite, ordinary MemoryAgent/store APIs, deterministic local generation fixtures, real parser regressions and owned child-process failure/restart tests. Counts include inherited tests, not independent benchmarks.

Compatibility and operations

New databases use the history-enabled schema. Existing v1 data requires explicit quiesced migration; earlier experimental unpublished v2 layouts require reviewed reconciliation rather than silent adoption. Historical raw APIs are trusted-host surfaces, and the integrated request facade must authorize them. History retirement is not physical erasure of source pages, backups or external copies. DELETE/FULL local SQLite is the tested path; no hardware-power-loss, live-provider or universal platform claim.

See docs/transactional-history.md and docs/closure/C03-interfaces.md for APIs, seven-query example, migration, bounds and recovery. C01/C02/C07/C09 final composition and release gates remain separate tracked work. No automatic merge.