refactor: remove the process-schedule remnant from the scheduler (tables created at boot, polled every sync, execute path posts to a route that no longer exists) #3432

Summary

The process engine was removed from the backend, but the scheduler still carries its scheduling half: it creates two tables at boot, re-reads them on every sync cycle, and holds an execute path that posts to a backend route that no longer exists. Nothing in the tree can create a row, so the code is dead by construction and still costs a query per sync cycle on every instance.

Context

Static review of dev at f6bedcd8. Sibling of #2491 (the non-scheduler process-engine remnants) and #2494 (memory-doc drift); kept separate because #2491 is deletions of unreachable files, while this removes branches from live scheduler code paths.

Why it is dead, not merely unused:

• process_schedule appears in exactly two production files, both under src/scheduler/ (service.py, database.py). The backend has no router, service, db module or table definition for it.
• The execute path posts to /api/processes/{id}/execute (src/scheduler/service.py). No such route exists.
• create_process_schedule, add_process_schedule, remove_process_schedule and update_process_schedule have no caller anywhere in src/.
• The two tables are created ad hoc by ensure_process_schedules_table. They are not in db/schema.py, db/tables.py or any Alembic revision, which the schema invariant forbids.
• Four process_schedule_execution_* WebSocket event types are published only from this path; no frontend listener references them.

Acceptance Criteria

• The scheduler no longer creates, loads, syncs or executes process schedules; git grep -n process_schedule src/ is empty
• ProcessSchedule and ProcessScheduleExecution are gone from src/scheduler/models.py
• The lock-denial and skipped-run helpers shared with agent schedules (bug: cron tick suppressed by distributed lock leaves no execution record #1969, bug(scheduler): _execute_process_schedule lock-denial branch still unaudited (sibling of #1969) #1994) keep working for agent schedules; only the process branch is removed
• tests/unit/test_1994_process_schedule_lock_audit.py is removed; the process cases in tests/scheduler_tests/ (test_skipped_executions.py, test_1474_timestamp_z_suffix.py, test_sync_loop.py) are cut and the agent-schedule cases still pass
• tests/run-core.sh no longer passes --ignore=process_engine (the directory is not in the tree)
• docs/memory/feature-flows/scheduler-service.md, docs/memory/requirements/scheduling.md and docs/memory/architecture/execution.md no longer describe process schedules as a working part of the scheduler (archive and release docs stay as they are)
• An instance that already has the two empty tables boots and schedules normally after the change
• Scheduler and unit suites green

Technical Notes

• Open choice for the implementer: drop the two empty tables (needs a migration on both the SQLite and Alembic tracks) or leave them. Leaving them is harmless once nothing reads them; say which was chosen in the PR.
• This touches the scheduler's sync loop and reload path, so sequence it after the 1.0 release is cut, not during release testing.