Skip loading extensions when instantiating benchmarks

[alejoe91]

With many cases and large recordings, loading everything in RAM can cause out of mem issues.

We don't really need loading extensions at this point

[zm711]

Related to this @alejoe91 / @samuelgarcia I think the real heavy extension is the waveforms right? I like to keep my waveforms around (really for provenance), but once I have my templates/PCA done, I don't really want to have to reload them. I can't think of a good mechanism to only autoload the "final" extensions (since that could probably be different for different people), so it would be nice if we could do that, but I guess the only way currently is to not load the extensions and then load them one-by-one?

[chrishalcrow]

Related to this @alejoe91 / @samuelgarcia I think the real heavy extension is the waveforms right? I like to keep my waveforms around (really for provenance), but once I have my templates/PCA done, I don't really want to have to reload them. I can't think of a good mechanism to only autoload the "final" extensions (since that could probably be different for different people), so it would be nice if we could do that, but I guess the only way currently is to not load the extensions and then load them one-by-one?

Agree with @zm711 - we do this in our lab all the time. My proposal: expand the load_extensions to accept lists of strings. E.g.

sa = si.load_sorting_analyzer(
    folder="my_sa", 
    load_extensions=["unit_locations", "templates", "spike_amplitudes"],
)

For another PR though!