feat: transparent leader hint reconnect on consume

[vieiralucas]

Summary

• When consume receives gRPC UNAVAILABLE with x-fila-leader-addr metadata, the client transparently reconnects to the leader node and retries once
• Max 1 redirect per consume call to prevent infinite loops
• UNAVAILABLE errors without the leader hint metadata are raised as RPCError as before

Test plan

• Verify existing integration tests still pass (enqueue/consume/ack, nack redeliver)
• Verify UNAVAILABLE without leader hint raises RPCError
• Verify UNAVAILABLE with leader hint reconnects to the indicated address
• Verify a second redirect on the retry raises RPCError (no infinite loop)

🤖 Generated with Claude Code

---

Summary by cubic

consume now auto-reconnects once to the leader on gRPC UNAVAILABLE when x-fila-leader-addr is present, rebuilding the stub with cached TLS credentials to keep the stream alive.

• New Features

  • Follow the leader hint once per consume; a second redirect raises RPCError.
  • UNAVAILABLE without the hint raises RPCError; Cancelled and NotFound unchanged.
  • Redirect logic runs in consume_with_redirect.

• Refactors

  • Cache TLS channel credentials in @credentials; added extract_leader_addr and made LEADER_ADDR_KEY private.
  • Inlined close and trimmed method docs to pass RuboCop’s 100-line class limit.
  • Condensed build_channel_credentials branching for brevity.

^{Written for commit 7601fb1. Summary will update on new commits.}

[cubic-dev-ai]

No issues found across 1 file