Allow run_sorter to accept dicts

[chrishalcrow]

Allows run_sorter to accept dictionaries of recordings, and return a dict of sorters.

Longer-term aim, make this workflow just work:

pp_recs = si.common_reference(rec.split_by(‘group’))
sortings = si.run_sorter(pp_rec)
analyzer = si.create_sorting_analyzer(sortings, pp_recordings)

Think this PR is minimally invasive. The old function run_sorter can now accept a dict. If it gets a dict, it calls run_sorter for each recording and saves in a folder

{folder}/
    spikeinterface_info.json
    splitting_key_0/
      {whatever run_sorter spits outs}
    splitting_key_1
        {whatever run_sorter spits outs}
    …

And spikeinterface_info.json looks like e.g.:

{
    "version": "0.102.4",
    "dev_mode": true,
    "object": "dict of BaseSorting",
    "dict_keys": [
        4,
        5
    ],
    "split_by_property": “group”
}

Tried a bigger refactor here: https://github.com/chrishalcrow/spikeinterface/tree/sort-by-group which avoids the recurrent logic here, but it was a mess. This seems a lot nicer.

[alejoe91]

@chrishalcrow maybe worth saving the dict_keys dtype in the spikeinterface_info since this gets lost when dumping to json

[samuelgarcia]

Thank you Chris, this is cool.

We also need to implement the spikeinterface.load that handle to open this new folder type.

We should discuss a bit more the "object": "dict of BaseSorting"
I add in mind something like "Group[Sorting]"

Also not sure to clearly understand why we need this "split_by_property" in the dict because the information is dump in all sub-folder in recording.json no ?

@chrishalcrow maybe worth saving the dict_keys dtype in the spikeinterface_info since this gets lost when dumping to json

Not sure I like this ugly hack in the format spikeinterface_info.json.
List of keys can be int or str no ? json handle it.
The list is in the dict to open back the sub folder in the good order and will converted to str for folder.
So I do not see why.

I would be happy to have a call.