fix: cancel the pending slot state timeout when a pool slot is shut down

[pjfanning]

Motivation

Every state transition in NewHostConnectionPool's slot state machine cancels the timeout of the state it leaves and, if the new state defines one, schedules a fresh materializer.scheduleOnce task for it. Those tasks are not bound to the stage lifecycle, so they keep running after the stage has stopped.

Slot.shutdown(), which postStop calls for every slot, closed the connection and ran state.onShutdown but never cancelled that timeout. A pool that stops while a slot is in a state with a finite timeout — waiting for a connection, for a response entity subscription, or for the keep-alive timeout — therefore leaves a scheduled task behind that keeps the slot, and through it the whole pool logic with its connections and request contexts, reachable until the timeout elapses. Firing does nothing useful either: the task only invokes an async callback on a stage that is already gone.

Modification

Cancel the current timeout at the start of Slot.shutdown(), like every state transition and the slot error path already do.

Result

Shutting a pool down no longer leaves scheduled state timeouts, and the pool becomes collectable as soon as it stops.

Tests

• New NewHostConnectionPoolSpec: the pool flow is driven with probes until a slot waits for a response entity subscription that never comes, then the pool is torn down. A test scheduler records the tasks scheduled with that state's (distinctive) timeout so the test can assert they end up cancelled. Reverting only the cancelCurrentTimeout() call makes it fail — the task is still live after the pool stopped.
• sbt "http-core/testOnly org.apache.pekko.http.impl.engine.client.pool.NewHostConnectionPoolSpec" — pass (1 test).
• sbt "http-core/mimaReportBinaryIssues" — pass.
• scalafmt --mode diff-ref=upstream/main --test — pass.

References

None — found while auditing src/main for resource leaks.