Generalizing the low-level data API for ephys data

[glopesdev]

Points to consider:

• Ephys data is timestamped with different clock at acquisition time
  • There is a global correspondence table between clocks that can be used to build an interpolator but would be nice if the architecture could support this seamlessly
  • Names of files are different since we don't have a chunk timestamp at acquisition time
• Ephys data is too big to load entire chunks (support for sub-chunking at read time would be useful)
  • Each chunk of neuropixel probe data is 81 GB
  • Even just one chunk of per-sample timestamps is ~800 MB
  • np.fromfile supports chunking, could also consider using memory mapping
• There are multiple devices inside the ephys headstage, how should we group them?
  • Consider if they should be different devices at the dataset level, or sub-devices inside a headstage device
  • Might be useful to keep the flat stream structure at the dataset level