=============
Minor updates:
- Convert custom units with UnitAdaptor
- Cache coefficients by dimension pairs for greater reuse
- Handle and test for duplicate data rows on read
- Improve error message if file data is missing timestep
- Improve DataArray.from_df validation and messages
- Use gzip compression for smaller parquet files
- Faster comparison of dim names using python set
Fixes:
- Debug messages with non-str interval names in IntervalAdaptor
- Store dimensions as CSV, special-case convention for intervals
- Handle raw Exception from xarray v0.10
- Ensure DataHandle.get_data returns DataArray named as the input spec