Add save_bboxes.to_netcdf() and load_bboxes.from_netcdf() for complete netCDF4 I/O support

[manashk29]

Currently the ethology I/O API is asymmetric:

• Load: from_files() supports COCO and VIA
• Save: to_COCO_file() supports COCO only

There is no way to save an ethology Dataset in a format that
preserves the full xarray structure and reload it without
repeating the full validation pipeline.

Users who want to use
netCDF4 must manually serialise/deserialise the dict attributes
(map_category_to_str, map_image_id_to_filename) because netCDF4
cannot store Python dicts natively.

User must write a 20+ line of undocumented boilerplate from scratch to serialise and deserialise.