feat: add background scheduler infrastructure (Phase 3.1)

[natgurlain]

Summary

• Add APScheduler-based background job system with 6 scheduled jobs: memory consolidation (interval), goal check (interval), calendar scan (interval), strategist tick (interval), daily briefing (cron), and evening review (cron)
• Add ConnectionManager WebSocket registry for broadcasting proactive/ambient messages to connected clients
• Wire scheduler into app lifespan (start after DB init, shutdown before close) and register WS connections in the chat endpoint

Details

New modules:

• src/scheduler/engine.py — init_scheduler() / shutdown_scheduler() using AsyncIOScheduler, all jobs gated by settings.scheduler_enabled
• src/scheduler/connection_manager.py — ConnectionManager with connect, disconnect, broadcast (auto-prunes dead connections)
• src/scheduler/jobs/memory_consolidation.py — Finds sessions updated within last hour and runs consolidate_session() (periodic catch-all alongside the ad-hoc trigger in ws.py)
• src/scheduler/jobs/{goal_check,calendar_scan,strategist_tick,daily_briefing,evening_review}.py — Stub jobs logging execution, to be implemented in later Phase 3 PRs

Modified:

• src/app.py — Scheduler init/shutdown in lifespan
• src/api/ws.py — ws_manager.connect() / disconnect() on WebSocket lifecycle

Tests (12 new):

• ConnectionManager: connect/disconnect tracking, broadcast delivery, dead connection pruning
• Scheduler engine: disabled returns None, enabled registers all 6 jobs, safe shutdown
• Memory consolidation job: processes recent sessions, skips old ones, continues on individual failure

Test plan

• All 144 tests pass (132 existing + 12 new)
• Start backend: logs show "Scheduler started with 6 jobs"
• Stub jobs log at configured intervals
• SCHEDULER_ENABLED=false → no scheduler starts
• WS connect/disconnect reflected in connection manager

🤖 Generated with Claude Code