Fix sparse template handling in kilosort reader for SI-exported Phy data

[ttngu207]

Summary

• Fixes incorrect spike_sites and best_channel values when reading SpikeInterface-exported Phy curations
• SpikeInterface's export_to_phy exports sparse templates by default (since v0.101.0)
• The existing code assumed dense templates (all channels), causing incorrect channel lookups

Problem

SpikeInterface exports templates.npy with shape (n_templates, n_samples, max_sparse_channels) where:

• max_sparse_channels is the maximum number of sparse channels across all units (not all channels)
• A companion file templates_ind.npy maps (template_idx, sparse_channel_idx) -> actual_channel_idx
• Padding positions use -1

The existing code used argmax to find the channel with max amplitude, but this returns an index into the sparse representation (0, 1, 2...), not the actual channel index.

Solution

Updated get_best_channel() and extract_spike_depths() to:

1. Check if templates_ind.npy exists (indicates SI-exported sparse format)
2. Use templates_ind to map sparse indices to actual channel indices
3. Fall back to original behavior for native Kilosort (dense) format

Test plan

• Test with SI-exported Phy curation data (MBA probes)
• Verify spike_sites values match expected channel indices
• Verify backward compatibility with native Kilosort output

Related

• Addresses Issue add get_sync_ephys_function for ephys sync & remove the irrelevant ephys modules #2 in dj-sciops/nei_nienborg#111
• Related Jira: SCIOPS-1434
• Companion to PR Fix KeyError in ApplyOfficialCuration.make for multi-insertion probes #19 (channel2electrode_map fix)

🤖 Generated with Claude Code