[jaideeppyne]

Fixes #182.

When an async RecvFut is polled after its hook has been fired but it fails to actually receive a message (e.g. the message was stolen by another receiver on a shared channel), the hook is re-inserted into Chan::waiting. The hook's woken flag was never cleared on re-insertion, so every subsequent poll that fails to receive re-observes woken == true and pushes yet another duplicate clone of the same hook, growing waiting without bound (a memory leak, and it corrupts the waiting-receiver accounting).

Interleaving, with two receivers sharing one channel:

1. R1 polls recv_async — hook H1 is registered in waiting, woken = false.
2. A sender pops H1 and calls fire(), setting woken = true; the message goes to the queue.
3. R2 (or a try_recv) steals that message first.
4. R1 is polled again: recv_sync fails, so H1 is re-inserted — but woken stays true.
5. Every later failing poll re-observes woken == true and pushes another copy of H1.

This clears the woken flag while re-arming and re-inserting the hook, under the channel lock. fire also runs under that lock and the hook is not in waiting during this window, so the reset is race-free. It only triggers on the path where recv_sync has already failed (queue empty), so no message is dropped. This is orthogonal to #177's missed-wakeup fix.

Includes a deterministic regression test: before the fix, 10 spurious polls leave 10 duplicate hooks in waiting; after, exactly 1. Full test suite passes across all features; fmt and clippy are clean.