Add graceful shutdown support (SIGTERM handling, readiness drop, gRPC GracefulStop) to avoid 500s during upgrades

[peresureda]

### Summary
We recently switched from Recreate to RollingUpdate (with a ConfigMap checksum annotation), and this successfully removed downtime for new connections. However, existing in‑flight requests still produce a small number of 500 errors during pod termination when running under Kubernetes with an Istio sidecar.

The reason is that the ratelimit binary does not implement graceful shutdown:

• It does not handle SIGTERM
• It does not mark readiness to false when shutting down
• gRPC is closed immediately instead of using GracefulStop()
• In‑flight requests are not drained
• Redis and internal workers stop abruptly

Even with Istio’s terminationDrainDuration and a preStop hook, the process exits too fast and causes 500s.

### Proposal
I would like to contribute upstream support for proper graceful shutdown:

• Catch SIGTERM
• Drop readiness immediately
• Use grpc.Server.GracefulStop()
• Allow in‑flight requests to finish (or timeout via env var)
• Close Redis and workers after draining

Are maintainers open to this contribution?

Thanks!