Fix #2068

[yhirose]

No description provided.

[falbrechtskirchinger]

The problem is that this now masks a clear programmer error. The assertion exists for a reason and should trigger reliably (so it can be seen and fixed).

I can see why users might expect stop() to wait, given that the library blocks in most other situations. stop() would be more accurately named request_stop(). Then, stop() (with a timeout) could wait and return false on timeout.

@yhirose Thoughts?

Ideally, we would add stop(void) calling stop(timeout=0) and mark the former as deprecated, but the library has so far avoided compiler-agnostic macros for such tasks.