Add use_times option in get_duration/get_total_duration

[alejoe91]

Since the __repr__ uses the get_durations function, I added an option to avoid using timestamps to estimate duration.
In fact, this can be quite slow for long zarr datasets with timestamps, since the whole array is loaded in mem when calling get_times().

[h-mayorquin]

Where do you think the loading into memory is happening?
Maybe here in the asarray:

https://github.com/h-mayorquin/spikeinterface/blob/80820853354fbc2c6a56bb5dade4c94c8dacb321/src/spikeinterface/core/baserecording.py#L895-L898

Because the duration only requires the last and the first timestamps:

https://github.com/h-mayorquin/spikeinterface/blob/80820853354fbc2c6a56bb5dade4c94c8dacb321/src/spikeinterface/core/baserecording.py#L237-L240

Maybe we can just not transform to array if the timestamps are memmap and or zarr provided they are read only?

My concern is that changing the repr with these defaults will display wrong information for some cases but maybe I am wrong.

[alejoe91]

I think that transforming to array is very convenient to cache the timestamps in memory and overall.it improves performance. Another option could be to have a get_start_time and get_end_time, which will only access the required samples.

[h-mayorquin]

Fair enough, changing the way that get_times work would probably require more discussion.

Adding private methods like this:

get_start_time and get_end_time, which will only access the required samples

Seems like a great idea to me. That way, we can use that in get duration without touching anything else.