Refactor channel inference in inferSubEpochChannels function to impro…

[yoshihito-saito]

Summary

This PR updates NeuroScope2 sub-epoch dat handling so that the channel count comes from session.extracellular.nChannels instead of being inferred from the timestamp-derived sample count.

Motivation

Some Intan datasets can have small timestamp gaps in time.dat. In those cases, MergePoints.timestamps_samples / session.epochs may report a sample span that is slightly longer than the actual number of samples stored in amplifier.dat.

For example, one dataset had:

• amplifier.dat: 157,831,296 actual samples at 128 channels int16
• MergePoints/session: 157,831,323 expected samples
• time.dat: timestamp jump from 152,339,682 to 152,339,711

The raw file size was valid for 128 channels, but channel inference failed because it used the timestamp-derived sample count.

Changes

• Treat session.extracellular.nChannels as the authoritative channel count for sub-epoch dat files.
• Validate that file size is divisible by nChannels * sampleBytes.
• Derive sub-epoch sample count from file size.
• Warn, rather than error, when metadata sample count differs from file-derived sample count.

Validation

• Confirmed NeuroScope2.m parses successfully with MATLAB mtree.
• Tested against the real failing dataset: