Force original_end_frame >= original_start_frame in tracepad

[TomBugnon]

Avoids this error in phase_shift (and possibly other places)

  File "/home/tbugnon/projects/shared_sortings/spikeinterface/src/spikeinterface/preprocessing/phase_shift.py", line 87, in get_traces
    traces_chunk, left_margin, right_margin = get_chunk_with_margin(
  File "/home/tbugnon/projects/shared_sortings/spikeinterface/src/spikeinterface/core/recording_tools.py", line 280, in get_chunk_with_margin
    traces_chunk2 = np.zeros((full_size, traces_chunk.shape[1]), dtype=dtype)
ValueError: negative dimensions are not allowed

[alejoe91]

@TomBugnon this error also appears for concatenated recordings, right?

[TomBugnon]

@alejoe91 I got this on a concatenated neuropixel recording (with full preprocessing)
I haven't tried on a simple recording, should I?

[alejoe91]

No no, I think it should work with simple data!

[samuelgarcia]

I have the feelign that this is hidding a deeper bug.
How we could have this.
Would it be better to raise an error instead no ?

[TomBugnon]

Here's a real life scenario where original_end_frame < original_start_frame in TracePad

#1881 (comment)

[TomBugnon]

(I do think there is an unrelated bug in get_chunk_with_margins , tho): #1885 (comment)

[h-mayorquin]

Sam wanted to look at this so I am just writing this for us as a reminder.

Also, we are discussing this issue in more generality here:
#1989
Feedback is welcomed.

[alejoe91]

@TomBugnon sorry for getting back so late on this.

This issue is gone with #1883 right? In that case I'll close this. Feel free to reopen if that's not the case!