Preserve client dispatch trace writes in Redis Cluster mode

[zhchxiao123]

What to build

Ensure Java by client message dispatch keeps writing client dispatch trace metadata, spans, and lookup indexes when the SDK is running against Redis Cluster. Message delivery should not lose observability data simply because the trace writer still assumes a standalone Redis resource.

The trace meta and span writes should use a Cluster-compatible access path and preserve the existing best-effort behavior for cross-entity lookup indexes: failures in one index write should not fail message dispatch or prevent TTLs on trace data that was successfully written.

Acceptance criteria

• Client dispatch trace meta and span writes succeed in Redis Cluster mode for the v2 trace key schema.
• Trace TTLs are applied to both trace metadata and span storage in Redis Cluster mode.
• Session, worker, and agent trace indexes remain best-effort independent writes and do not fail message dispatch.
• Existing standalone trace write behavior remains backward compatible.
• Tests cover Cluster-mode trace writing and verify no direct standalone-only Redis resource is required.

Blocked by

• Enable Redis Cluster initialization for Java client #31