[hub] Fix stale abort listener terminating pooled sha256 workers - #2325

Fixes #2311.

The abort listener in the pooled web-worker path is only removed when it fires, so a successful hash leaves it attached to the caller's signal. When that signal aborts later — one signal usually covers a whole multi-file commit, so this is the ordinary case — the listener runs against an operation that already finished and:

• terminates a worker that freeWorker already returned to pendingWorkers
• rejects a generator that already returned, producing an unhandled AbortError

cleanup() now removes the abort listener too, so every exit path — success, worker error, non-hash message, abort — drops it in one place.

The test drives the pooled path with a stand-in Worker (Node has no global one), completes a hash, then aborts. Without the fix it fails on the terminate count and throws an unhandled ABORT_ERR. It's excluded from the browser config for the same reason uploadShards.spec.ts is — vi.mock.

One thing I noticed but left alone: destroyWorker only deletes from runningWorkers, so a worker terminated while sitting in pendingWorkers stays in the pool and getWorker hands it to the next caller. Nothing reaches that state with this fix in place, but happy to harden it separately if you'd like.