[KIP-932]: Per partition error handling tests and per RPC error handling tests

[Pratyush Ranjan (PratRanj07)]

Per-partition error handling tests (0182-share_consumer_error_handling_mock)

• test_partition_error_injection_share_ack_matrix / do_test_partition_error_injection_share_ack_code — a per-partition ShareAck err propagates to commit_sync results and the ack callback for the affected partition only; other partitions stay NO_ERROR.
• test_partition_error_injection_share_fetch_surfaces_err — a per-partition ShareFetch err reaches the app via share_poll while records from unaffected partitions are still delivered.
• test_partition_error_injection_per_apikey_stack_isolation — ShareFetch and ShareAck error stacks are scoped per ApiKey and drain independently (no cross-contamination).
• do_test_share_fetch_partition_err_triggers_metadata_refresh — leader-unavailable-class errs (NOT_LEADER_OR_FOLLOWER, FENCED_LEADER_EPOCH, KAFKA_STORAGE_ERROR, OFFSET/REPLICA_NOT_AVAILABLE) trigger a Metadata refresh and recover transparently (err never surfaces).
• do_test_share_fetch_partition_err_silent_await / test_share_fetch_partition_err_silent_await_matrix — unknown-topic-class errs (UNKNOWN_TOPIC_OR_PART, UNKNOWN_TOPIC_ID, INCONSISTENT_TOPIC_ID) don't surface; records arrive once the err stack drains.
• test_share_fetch_partition_err_default_translates_to_state — an unmapped ShareFetch err code is translated to __STATE before surfacing; the raw broker code is never visible to the app.
• test_share_fetch_partition_err_unknown_leader_epoch_log_only — UNKNOWN_LEADER_EPOCH is log-only, doesn't surface, and records deliver after the stack drains.
• test_share_fetch_partition_err_on_subsequent_fetch_recovers — a mid-session leader-unavailable err on a later ShareFetch still triggers a Metadata refresh and recovers.
• test_share_ack_partition_err_preserves_session — a failed per-partition ack does not reset the share session; the next fetch+ack+commit cycle is clean with no extra rejoin.
• test_share_ack_partition_err_not_auto_retried — the client never silently retries a failed ack; a follow-up commit_sync with no new acks sends zero ShareAcknowledge on the wire.
• test_share_ack_partition_err_after_clean_ack_surfaces — an ack err on a second commit_sync (after a clean cycle) propagates to the matching partition's result and callback only.
• test_share_group_adherence_to_hb_interval — ShareGroupHeartbeats fire at the configured interval (3–5 over ~2s at 1000ms).
• test_share_group_metadata_unknown_topic_id_tests — a topic carrying UNKNOWN_TOPIC_ID in Metadata can't be fetched until cleared, while a clean co-subscribed topic still delivers.
• do_test_share_group_quick_unsubscribe / test_share_group_quick_unsubscribe_tests — rapid subscribe/unsubscribe cycles trip no assertion and destroy cleanly, both with a ready cluster and against an unreachable bootstrap.
• test_share_fetch_fast_leader_query_backoff — a NOT_LEADER_OR_FOLLOWER on ShareFetch triggers a fast-leader-query Metadata request right after the failing fetch.

[confluent-cla-assistant]

🎉 All Contributor License Agreements have been signed. Ready to merge.
_{Please push an empty commit if you would like to re-run the checks to verify CLA status for all contributors.}