fix: resolve idempotency middleware and health check issues

[MatanOga01]

Summary

This PR resolves all issues assigned to MatanOga01:

Closes

• Closes Idempotency middleware caches error responses: a failed request is replayed as success for 24 hours ApexChainx/ApexChainx-Backend#312
• Closes Idempotency keys are not scoped to the authenticated user: one user's cached response replays to another ApexChainx/ApexChainx-Backend#313
• Closes Idempotency middleware 500s when Redis is down: every state-changing request with a key fails hard ApexChainx/ApexChainx-Backend#314
• Closes Health check catches redis.ConnectionError around a Postgres probe: DB outages never mark the app unhealthy ApexChainx/ApexChainx-Backend#315

Changes

ApexChainx#312 — Idempotency middleware no longer caches error responses

• Only 2xx responses are stored in Redis; 5xx responses evict the key so retries can re-execute
• Prevents a failed first attempt from poisoning the key for the full TTL

ApexChainx#313 — Idempotency keys are now scoped per user

• Cache key now includes a hash of the bearer token: idempotency:{actor_key}:{idempotency_key}
• Prevents cross-user response replay when two users send the same Idempotency-Key

ApexChainx#314 — Idempotency middleware no longer 500s when Redis is down

• Added try/except around all Redis operations with a circuit breaker (30-second trip)
• Fails open: requests are processed without caching when Redis is unavailable
• Logs a warning on each circuit trip

ApexChainx#315 — Health check now catches the correct exception type

• Removed the dead check_database() / check_celery() helpers from app/main.py that incorrectly caught redis.ConnectionError around a Postgres probe
• The live /health/readiness endpoint already uses build_readiness_report() in health_report.py which catches sqlalchemy.exc.OperationalError correctly

Files changed

• app/middleware/idempotency.py
• app/main.py