Add option to export spike waveforms subset to phy (mimic phy extract-waveforms)

[alejoe91]

Summary

• Adds an add_waveforms option to export_to_phy that writes phy's _phy_spikes_subset.{spikes,channels,waveforms}.npy files, mimicking the phy extract-waveforms CLI command.
• Reuses the SortingAnalyzer's waveforms/random_spikes extensions when already computed, and computes them on demand otherwise (skipping with a warning if there's no recording available).
• Handles both sparse and dense SortingAnalyzer/sparsity combinations, matching the sparsity already used for templates.npy/pc_feature_ind.npy.

Test plan

• Verified with phylib.io.model.TemplateModel (phy's own loader) that the exported files load correctly and return proper per-cluster waveforms, for both sparse and dense analyzers.
• Verified stored waveform values exactly match manually-extracted raw-trace snippets (accounting for nbefore/uV scaling).
• Add a unit test to src/spikeinterface/exporters/tests/ covering add_waveforms=True.

🤖 Generated with Claude Code

[chrishalcrow]

I think the new release of phy no longer supports _phy_spikes_subset.* - I can't find any mention of this type of file in the code of the codebase. I propose to delay this for now.

[alejoe91]

I think the new release of phy no longer supports _phy_spikes_subset.* - I can't find any mention of this type of file in the code of the codebase. I propose to delay this for now.

This is handled in phylib and I tested with the most recent phy version:
https://github.com/cortex-lab/phylib/blob/master/phylib/io/model.py#L684-L702

Example with add_waveforms=True, exported after resetting recoriding (so waveforms must be coming from new files), phy==2.1.0

I don't think we should delay this ;)

[chrishalcrow]

Oh, there was a bug(?) in phy and no templates or waveforms were showing until I toggled and bunch of options. Ok, working for me now!

Just one request to make one of the docstrings less overly verbose.