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

[manashk29]

Hii @sfmig
Closes #152

What is this PR

• Bug fix
• Addition of a new feature
• Other

Adds two new public functions to complete the netCDF4 I/O round-trip
for ethology bounding box annotation datasets.

##save_bboxes.to_netcdf()

What it does:
Wraps xarray.Dataset.to_netcdf() with automatic handling of all
attribute types that netCDF4 cannot store natively.

The problem it solves:
The ethology Dataset has three dict attributes
(map_category_to_str, map_image_id_to_filename,
map_image_id_to_original_coco_id), a list attribute
(annotation_files), and a None attribute (images_directories).
netCDF4 cannot store any of these types — calling ds.to_netcdf()
directly would either silently corrupt them or raise a TypeError.

How it works:

1. Makes a deep copy of the Dataset so the caller's attrs are never mutated
2. Iterates over all attrs and converts:
   • dict → JSON string via json.dumps()
   • list → JSON string (each element converted to str first)
   • Path → plain string
   • None → dropped entirely
   • str, int, float → kept as-is (already netCDF4-safe)
3. Calls xarray.Dataset.to_netcdf() on the cleaned copy

##load_bboxes.from_netcdf()

What it does:
Wraps xr.open_dataset() with automatic deserialisation of all
attributes back to their original Python types, then validates
the loaded Dataset.

The problem it solves:
After loading a netCDF4 file with raw xr.open_dataset(),
map_category_to_str comes back as the string
'{"1": "Mallard", "3": "Goose"}' instead of the dict
{1: "Mallard", 3: "Goose"}. The keys are also strings ("1", "3")
not integers (1, 3) because json.dumps() always serialises dict
keys as strings. Any downstream code calling
ds.attrs["map_category_to_str"][1] would fail with a KeyError.

How it works:

1. Raises FileNotFoundError with a clear message if the file does not exist
2. Calls xr.open_dataset(file_path).load() — the .load() call
   reads all data eagerly into memory and closes the file handle
   immediately, preventing PermissionError on Windows
3. Iterates over the three dict attrs and for each:
   • Calls json.loads() to parse the JSON string back to a dict
   • Converts all numeric string keys ("1", "3") back to integers
     (1, 3) using int(k) if k.lstrip("-").isdigit() else k
4. Restores annotation_files from JSON string back to a list of
   Path objects
5. Uses @_check_output(ValidBboxAnnotationsDataset) decorator —
   same as from_files() — so the returned Dataset is guaranteed
   to be valid

Tests

Added tests/test_unit/test_io_annotations/test_io_netcdf.py with 3 test classes

Checklist:

• The code has been tested locally
• Tests have been added to cover all new functionality
• The documentation has been updated to reflect any changes
• The code has been formatted with pre-commit