[TheNetherWatcher]

check_sig_cohesion finds the minimum and maximum of each channel to check that the digital samples fit the format. It used Python's built-in min and max, which go through a NumPy array one element at a time, so this check took about 85% of the time of wrsamp for large records. This uses np.min and np.max instead, in both the single-frequency and the expanded (multi-frequency) path.

Time for a 4-hour, 8-channel record at 500 Hz (57.6 million samples), written from digital samples:

Format	check_sig_cohesion before → after	wrsamp before → after
16	2.41 → 0.07 s	2.85 → 0.45 s
212	2.41 → 0.08 s	3.07 → 0.78 s

The written files are unchanged. There were no tests for this check, so I added TestSigCohesion: a value just outside the range in the last sample is rejected (in both paths), and the smallest and largest values of each format are allowed. These pass on main as well, and fail if the check skips the last sample.

Related: #607 speeds up the byte packing in the same wrsamp call. The two don't touch the same lines, apart from both adding a changelog entry and a test class at the end of the file.

pytest -m "not network" passes (101 tests), and the docs build passes.

[tompollard]

Thanks, this looks good, and it's nice to have tests for this check now! For a 1-hour, 8-channel record the check went from 0.61 s to 0.02 s on my machine.