fix: call deduplicateCallback on topics deletion error

[kjamrog]

What

Fix the error path of Admin#deleteTopics to call deduplicateCallback instead of the raw callback.

Why

On error, #deleteTopics invoked the outer callback directly and never called deduplicateCallback. Because kPerformDeduplicated only clears its in-flight entry (and flushes queued callers) when that completion callback fires, a failed deleteTopics left a stale entry in the deduplication cache. As a result, any subsequent deleteTopics for the same topic set on the same Admin instance would be queued forever and never resolve — and concurrent duplicate callers would hang on the first failure too.

A common trigger is deleting a non-existent topic (e.g. test cleanup): the first call errors and poisons the cache, so every later call hangs.

This aligns #deleteTopics with #createTopics, which already calls deduplicateCallback(new MultipleErrors(...)) on error.

Consequences

• Dedup cache is reset on failure, so retries/subsequent calls work instead of hanging.
• All deduplicated callers reject with the same error rather than hanging.
• Single-call error output is unchanged (MultipleErrors('Deleting topics failed.', [error])); existing tests still pass.

Test

Adds a regression test asserting that a second deleteTopics for the same topics still rejects (rather than hangs) after a prior failure.