feat: add graceful shutdown with drain timeout and per-RPC timeouts

Graceful shutdown (Phase 6.4):
- Listen for SIGTERM + SIGINT via tokio::signal
- Configurable drain timeout (--drain-timeout / QPQ_DRAIN_TIMEOUT, default 30s)
- Health endpoint returns "draining" during shutdown for load balancer awareness
- ServerState carries atomic draining flag
- Add RpcStatus::Unavailable (9) for shutdown-related rejections

Per-RPC timeouts (Phase 6.5):
- Add RpcStatus::DeadlineExceeded (8) for server-side timeouts
- MethodRegistry supports default_timeout and per-method timeout overrides
- RPC dispatch wraps handler invocation with tokio::time::timeout
- RequestContext carries optional deadline (Instant) for handlers
- Health: 5s timeout, blob upload/download: 120s timeout, default: 30s
- Config: --rpc-timeout / QPQ_RPC_TIMEOUT, --storage-timeout / QPQ_STORAGE_TIMEOUT

crates/quicproquo-rpc/Cargo.toml

@@ -18,6 +18,8 @@ tracing          = { workspace = true }
 thiserror        = { workspace = true }
 dashmap          = { workspace = true }
 sha2             = { workspace = true }
+uuid             = { version = "1", features = ["v7"] }
+metrics          = "0.22"
 
 [dev-dependencies]
 tokio = { workspace = true, features = ["test-util"] }