Bug in extract_waveforms_to_buffers()

Using extract_waveforms_to_buffers() I noticed there is a mismatch between the number of expected waveforms and the actual number returned per unit, when the unit_ids not exactly like range(num_units).

It seems my issue is related to this piece of code in allocate_waveforms_buffers():

for unit_ind, unit_id in enumerate(unit_ids):
        n_spikes = np.sum(spikes["unit_index"] == unit_ind)

https://github.com/SpikeInterface/spikeinterface/blob/2b6e7a2ed43f946274209b0fa7d63e96e0be2f55/src/spikeinterface/core/waveform_tools.py#L183

Obviously, if unit_ids=[0,1,2,3,4] there is no issue, but if unit_ids=[0,1,13,91,101], for example, then there is an issue, because:

>>> unit_ids=[0,1,13,91,101]
>>> list(enumerate(unit_ids))
[(0, 0), (1, 1), (2, 13), (3, 91), (4, 101)]

I could be missing something about how extract_waveforms_to_buffers() is intended to be used, but to me this looks like it was a mistake.

Start in src/spikeinterface/core/waveform_tools.py at allocate_waveforms_buffers(), especially the loop around line 183, and trace how unit_ids are used by extract_waveforms_to_buffers(). Reproduce the mismatch with non-contiguous IDs such as [0, 1, 13, 91, 101], then verify that the returned waveform count for each unit matches its actual spikes.