refactor(db): split db/schedules.py into a 10-mixin package (#1481)

[AndriiPasternak31]

Fixes #1481

Summary

Pure mechanical split of the 4,139-line src/backend/db/schedules.py into a 10-mixin db/schedules/ package composed into ScheduleOperations — the sanctioned db/agent_settings/ shape (Invariant #2). The facade import path from db.schedules import ScheduleOperations is preserved by the package re-export, so database.py and db/__init__.py are unchanged. Zero behavior change, zero schema change, zero WHERE-clause edits, zero signature edits.

Mixins: ScheduleCommonMixin / Crud / Webhooks / Executions / Queue / Cleanup / Analytics / Stats / GitConfig / Retention. No import edges between mixin files — every cross-slice reference resolves via the composed class's MRO (self.<method>()); the only intra-package import is the bare _norm_ts module-global from _common.py.

Proof (mechanical-move invariance)

• 103/103 function bodies AST-byte-identical vs pre-rebase dev (symbol-set diff: 103 functions, none added / dropped / renamed; AST source-segment body diff: all 103 bodies byte-identical).
• 20 status-CAS writers (refactor: status-as-projection — schedule_executions.status never read as authoritative for "is running" #1082) re-discovered by a package-wide AST scan — the writer-inventory tripwire now rglob-scans the whole db/schedules/ package (was one hardcoded file) with an in-function assert sites non-empty guard so the CAS-precondition check can never pass vacuously; found == _EXPECTED_UPDATE_SITES both ways.
• 3 fat signatures untouched (create_task_execution / update_execution_status / create_schedule_execution) — byte-identical, reserved for refactor(db): reduce database.py facade signature width — param objects for execution/chat writers #1482.
• Only allowlisted text deviations: relative-import depth (.→..), the getLogger("db.schedules") literal pinned in the 3 logging modules, and the _PERCENTILE_ROWSET_CAP monkeypatch doc-comment retargeted to db.schedules.analytics.

Lockstep test edits (the only non-move work): writer-discovery rglob rework + db.schedules.* sys.modules eviction widening (conftest + test_agent_analytics.py) + _PERCENTILE_ROWSET_CAP monkeypatch retarget (module-identity trap).

Verification

• Crux test tests/unit/test_schedule_status_observability.py: 19 passed, 1 skipped; the writer inventory found exactly _EXPECTED_UPDATE_SITES (20 update sites) with no missing / stale entries.
• Lockstep-edited test_agent_analytics.py + test_schedule_analytics.py: 25 passed.
• Import smoke: from db.schedules import ScheduleOperations resolves cleanly (95 methods).
• verify-local --skip-agent: PASS including import-smoke, proving the db/schedules/ package ships in the prod image (docker/backend/Dockerfile COPYs db/ wholesale). 1 red = pre-existing test_1474 on dev, proven identical on base (not introduced by this PR).

Follow-ups (for Andrii)

• docs/memory/feature-flows/scheduling.md refresh — path/line refs to db/schedules.py were already stale pre-PR; ~30 flow docs cite the physical path (the import path is unchanged).
• F841 dead FAILED_STATES in the analytics slice (analytics-dedup).
• Stale db/schedules.py prose comments referencing the old single-file layout.

Sequencing: #1482 (facade param-objects) rebases on this.

🤖 Generated with Claude Code

[github-actions]

⚠️ Nightly unit-suite check skipped — merge conflict against dev.

Resolve by running git merge dev locally and pushing the result. The next nightly run will re-test once the conflict is gone.

[vybe]

Validated via /validate-pr: mechanical split with facade re-export preserved, 103/103 AST-byte-identical bodies, writer-inventory tripwire now rglob-scans the package (no vacuous pass), all checks green incl. regression-diff. No schema change, no packaging impact (db/ COPYed wholesale, prod-image-smoke green). Approving.