- Datasets are now loaded lazily. This should increase performance when opening
files with a large number of groups and/or variables.
- Support for writing arrays of variable length unicode strings with
``dtype=str`` via the legacy API.
- h5netcdf now writes the ``_NCProperties`` attribute for identifying netCDF4
files.