Add graceful degradation for external service dependencies

[railway-app]

Problem

When Postgres is unavailable at startup the application crashes or enters a boot loop, and when Redis is down the health endpoint returns 500 instead of a meaningful degraded status. There is no structured startup logging to show which services are reachable, making it hard to diagnose transient dependency failures.

Solution

Added a lazy DB availability probe (probe_db / is_db_available) in db/session.py that never raises — the engine object is created unconditionally (SQLAlchemy defers the TCP handshake to the first query) and a cheap SELECT 1 is used to detect reachability. The lifespan in main.py now probes Postgres, Redis, and Metrics at startup and emits structured [STARTUP] Service: STATE (note) log lines followed by a summary line (Application ready (all services healthy) or Application ready (degraded mode)). The root /health endpoint now returns 503 with "postgres": "unavailable" when the DB probe fails, and the deep /api/v1/health endpoint returns 503 when Postgres is down while keeping 200 for Redis/Discord failures (non-critical). The _boot_guard middleware is unchanged — it only blocks on hard config errors, not DB unavailability, so the app continues to serve all routes in degraded mode.

Changes

• Modified backend/app/db/session.py
• Modified backend/app/api/v1/endpoints/health.py
• Modified backend/app/main.py

---

Generated by Railway