CI gate: fail PRs that add a migration without updating schema.py

[AndriiPasternak31]

Context

Follow-up to #691 / PR #712.

#691 backfilled src/backend/db/schema.py with 11 tables, 14 columns, and ~22 indexes that existed only in src/backend/db/migrations.py. The fix re-encoded the drift class — there is no automated check preventing the next contributor from adding a migration without updating schema.py. In 6 months, the same drift will accumulate again unless we close the loop.

This was raised by both Codex and an independent Claude subagent during the autoplan dual-voice CEO review for #691, and approved at the final approval gate (D2 = B) as a follow-up issue to be filed alongside the implementation PR.

Proposal

Add a CI check that runs the parity verification scripts already documented in #691's plan against every PR that touches src/backend/db/migrations.py:

1. Spin up an in-memory SQLite database, run init_schema() only, snapshot state.
2. Run run_all_migrations(), snapshot state again.
3. Diff tables, columns, indexes. Filter schema_migrations (created internally by run_all_migrations).
4. Run a strict DDL parity check: normalize whitespace and lowercase the sqlite_master.sql text, compare entry-by-entry. Fail on any MISSING or DIFFERENT line.

Implementation sketch

A small Python check at scripts/check-schema-parity.py (or tests/unit/test_schema_parity.py so it runs with the existing test suite). The verification script from #691's plan is essentially the spec — it works against the current branch (verified during autoplan) and produces empty output when schema.py is faithful.

Wire into CI by:

• Running it as part of tests/run-core.sh (Tier 2 pre-commit checks)
• Adding a GitHub Actions job that runs only when src/backend/db/migrations.py or src/backend/db/schema.py changes (path filter)

Acceptance criteria

• Script lives at a discoverable path and exits non-zero on drift
• CI runs the script on every PR that touches migrations.py or schema.py
• Adding a migration without updating schema.py produces a CI failure with a clear "schema.py drift detected" message naming the missing table/column/index
• The check runs against the same dual-pass behavior as init_database() so it stays consistent with production semantics

Out of scope

• Collapsing the two-source-of-truth pattern (e.g., migrating to Alembic, generating schema.py from migrations) — separate, larger discussion.

References

• Schema Validation: 11 tables + 14 columns missing from schema.py (invariant #3 violated) #691 (the issue this prevents from recurring)
• PR fix(schema): backfill schema.py to match migrations.py reality (#691) #712 (the backfill that made schema.py faithful again)
• docs/memory/architecture.md — Architectural Invariant Feature/vector log retention #3

[AndriiPasternak31]

/claim

[github-actions]

Claimed by @AndriiPasternak31. Label status-in-progress added.

Next steps:

1. Create branch: feature/713-<short-slug>
2. Implement the feature
3. Open a PR with Fixes #713

[vybe]

Fixed in PR #745 (99d633e on dev). CI gate added — PRs that add a migration without updating schema.py now fail automatically.