refactor(compression): drop the terminate flag from pool removal

[anhtahaylove]

Follow-up to #12813 — please merge that one first, as this branch is stacked on it.

Rationale

After the idle-eviction fix, all three call sites pass terminate: true:

Call site	Argument
close()	true
finish() idle timer	true (was false — the bug in #12813)
fail()	true

The parameter now only exists to express a state the pool cannot safely be in: a slot deleted from this.workers while its Worker keeps running. Once remove() has done this.workers.delete(slot), the pool has dropped its only reference — nothing can terminate that thread afterwards. Passing false is not a valid option, it is a leak.

Removing the flag makes that state unrepresentable, so the same bug cannot be reintroduced by a future caller.

Change

-  private async remove(slot: PoolWorker, terminate: boolean): Promise<void> {
+  private async remove(slot: PoolWorker): Promise<void> {
     if (!this.workers.delete(slot)) return;
     if (slot.timeout) clearTimeout(slot.timeout);
     if (slot.idle) clearTimeout(slot.idle);
-    if (terminate) await slot.worker.terminate().catch(() => undefined);
+    await slot.worker.terminate().catch(() => undefined);
   }

Plus the three call sites dropping their now-redundant argument. remove() is private, so this is not a public API change.

No behavioural change on top of #12813.

Note on the sibling pools

I checked the other two worker owners in the codebase while investigating:

• open-sse/services/compression/engines/llmlingua/worker.ts — resetWorker() always calls w.terminate()
• src/lib/usage/callLogArtifactWriter.ts — terminateWorker() always calls .terminate()

Both are singletons that route every exit path (idle, error, exit, close) through a single teardown function, and both null out the reference before terminating. Neither has an equivalent flag, which is why neither leaked. compressionWorkerPool.ts was the only place where "remove from pool" and "stop the thread" were decoupled — this PR closes that gap.

[JasonBroderick]

Note for the maintainer: #12542 already removes the terminate parameter as part of the fix itself (with a regression test), so if #12542 is merged this follow-up becomes unnecessary. Field measurements for the fix are on #12812 and #12542.

[anhtahaylove]

Superseded by #13091, which carries the identical source change plus a regression test and a changelog entry. Closing this one to keep the review queue unambiguous — no need to review both.