Lazy-mode for loading Sortinganalyzer

[alejoe91]

The SortingAnalyzer, especially for chronic data, can be very memory hungry. This PR adds a lazy mode at load that:

• Reads the spike vector in lazy mode (memmaps memory folder, virtual memmap for zarr - TODO)
• Reads vector-extensions as memmaps or keeps them as zarr array
• Implements a memmap-like ZarrSpikeVector

NOTE: in lazy mode, save/select/merge/split operations are not permitted!

[chrishalcrow]

Hello, when I do a get_data() on any of the spike-vector extensions (e.g. spikeamplitudes) on a zarr'd analyzer I get an error here:

spikeinterface/src/spikeinterface/core/analyzer_extension_core.py

Line 1622 in ecf5c6d

return all_data.copy() # return a copy to avoid modification

as follows:

Traceback (most recent call last):
  File "<python-input-10>", line 1, in <module>
    sa.get_extension('spike_amplitudes').get_data()
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
  File "/home/nolanlab/fromgit/spikeinterface/src/spikeinterface/core/sortinganalyzer.py", line 3465, in get_data
    return self._get_data(*args, **kwargs)
           ~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
  File "/home/nolanlab/fromgit/spikeinterface/src/spikeinterface/core/analyzer_extension_core.py", line 1622, in _get_data
    return all_data.copy()  # return a copy to avoid modification
           ^^^^^^^^^^^^^
AttributeError: 'Array' object has no attribute 'copy'

[alejoe91]

.get_data()

So when it's lazy we should force copy=False. Does it make sense?

[chrishalcrow]

Works for me, and doesnt't change any defaults.
Note: it's pretty conservative. If you use lazy mode, you can't save/recompute/anything!

[samuelgarcia]

I would have been very happy to discuss this PR in detail before the merge
For perf reason mainly.
The ZarrSpikeVector would have deserve some deep benchmark no ?