But stopWaveform() refuses to do anything about it if the waveform was previously stopped by expired runtime from inside ISR, either. Fix for #7230 from original issue reporter.

For some reason stopWaveform() is in IRAM cache and optimized for performance over size.
Also, stopWaveform() would be explicitly called (if indirectly) from user code, which for waveform with a runtime argument can be reasonably expected to be well adjusted in time to avoid race conditions between ISR-based cutoff vs. stopWaveform().

Both facts talk back to #7232 (comment), which curiously mentions that 10µs isn't a big thing - for a function that bloats under gcc -O2 and is in IRAM cache :-)