fix: prevent stopped event buses from restarting a busy loop
Fixes #38.
A late dispatch still raises QueueShutDown, but _start() rejects the shut-down queue before creating another run loop. Shutdown and cancellation reach the loop's existing exit handlers, and the temporary queue-get task is always cleaned up.
Idle waits return after stop without restarting the bus, including waits already in progress. They use the existing idle signal, which the run loop sets after draining work and stop() sets when aborting it. The queue and event state are checked again before returning while the bus is running.
Reproduced the report's stop/dispatch sequence on Windows with Python 3.12: the original process stayed running after cancellation and used about one CPU core; the fixed version stays stopped and exits normally. Also reproduced a pending idle wait hanging during immediate stop on both the unchanged upstream base and the earlier PR head; it now returns.
Validation: nine shutdown/idle-wait cases pass, including a normal wait that must not return before its handler finishes. The full suite passes (147 tests, one existing duplicate-handler warning), as do the repository's pre-commit checks. The initial four shutdown regressions failed on the unchanged upstream base; two concurrent-stop regression cases failed on the previous PR head.