[server] Fix table deletion stuck when StopReplicaRequest send fails

[lilei1128]

Purpose

Linked issue: close #xxx

When a StopReplicaRequest send fails with a network-level throwable,
the coordinator had already transitioned replicas to ReplicaDeletionStarted
but silently dropped the error. This permanently stuck table deletion because:

• isEligibleForDeletion() blocks tables with any replica in ReplicaDeletionStarted
• resumeDeletions() only triggers from processDeleteReplicaResponseReceived,
  which is only reached when a DeleteReplicaResponseReceivedEvent is emitted
• processDeadTabletServer() skips replicas that are toBeDeleted

Brief change log

Fix: when sendStopRequest() fails with a throwable, emit a
DeleteReplicaResponseReceivedEvent with error results for the
delete-flagged buckets (delete=true && deleteRemote=true). This feeds
into the existing retry/give-up logic in processDeleteReplicaResponseReceived,
which retries up to DELETE_TRY_TIMES (5) and then force-marks replicas as
ReplicaDeletionSuccessful, allowing deletion to complete.

Non-deletion stop replicas (delete=false) continue to silently ignore
send failures as before, since they are best-effort.

Tests

Add testDeleteTableWithSendFailure() to verify that deletion completes
even when all StopReplicaRequest sends fail at the network level (i.e.,
CompletableFuture.failedFuture, not a response-level error code). The
existing testDeleteReplicaStateChange() only covered response-level errors.

API and Format

Documentation

[lilei1128]

close #3357