[kafka] Integrate Produce with native Fluss append

### Purpose

Part of #4185. PR07 of the Kafka compatibility sequence.

Wire the Produce handler and raw/string transcoder to the TabletServer append gateway. Apply the same table admission contract as Metadata, isolate conversion and append errors per partition, map acknowledgements and offsets, and register Produce v3-v11.

### Acceptance criteria

- [ ] Only compatible pre-created tables and valid bucket IDs reach append.
- [ ] One invalid table or partition does not suppress independent valid writes.
- [ ] Gateway acks/timeout and partition offsets/replication errors map correctly.
- [ ] Kafka calls tryCompleteActions immediately after synchronous produceLog invocation, including throws, without waiting for async acknowledgements.
- [ ] A real Kafka producer discovers native DDL tables and writes raw/string records with acks=0/1/all; native Fluss readback verifies data and metadata.

### Dependencies

PR06 raw/string conversion (#4282) and independent core PR #3412 / #3455. Core changes remain owned by #3412.

### Scope boundary

No new core implementation, Kafka table creation, JSON, authentication, idempotence or transactions.

### Implementation

Draft PR #4286: https://github.com/apache/fluss/pull/4286

Java 11 validation: 82 unit tests and 4 integration tests passed at `dc5fb6599b8c5e155cb17280df71a74ee34c9ffa`. Review and merge remain pending.

Start with the Produce handler, raw/string transcoder, and TabletServer append gateway described in the issue; compare draft PR #4286 and its listed validation commit, including Metadata's table-admission contract. Done means the acceptance criteria pass, including isolated partition errors, mapped acknowledgements and offsets, immediate completion behavior, and real Kafka/native Fluss integration validation.

Written by the indexing model from the issue text.