Fix infinite loop on broken ALSA pipes

[cdellacqua]

When an ALSA interface ends up in an unrecoverable error state (e.g. device disconnected), the thread that runs input_stream_worker (or output_stream_worker depending on the type of stream) goes into an infinite loop due to the polling always returning an error.

This infinite loop causes calls to drop(stream) to hang indefinitely as the Drop implementation tries to join the polling thread.

Proposed solution

By adding a flag that's raised when a Stream is dropped, we can add an early exit condition to poll_descriptors_and_prepare_buffer that forces a return value of PollDescriptorsFlow::Return.

Testing

I successfully tested the fix on my local machine by unplugging the microphone while listening to it.

[cdellacqua]

Note: the failed pipeline is wasm/emscripten-specific. The error looks quite similar to the one described in issue #810 but it's out of scope for this PR