Support opening EnMAP product directories (not only .tar.gz archives)

[remoteJorge]

Describe the bug
The reader works when opening .tar.gz product archives, but fails when opening an already-extracted product directory.
read_dataset_from_directory assumes files are named SPECTRAL_IMAGE.TIF, QL_PIXELMASK.TIF, etc., while in extracted products they keep the EnMAP prefix (e.g. ENMAP01-…-SPECTRAL_IMAGE.TIF).

To reproduce

# Works with .tar.gz
ds = xr.open_dataset(
    "/path/to/ENMAP01-____.tar.gz",
    engine="enmap"
)

# Fails with already-extracted directory
ds = xr.open_dataset(
    "/path/to/ENMAP01-____L2A-DT0000084125_20240719T111058Z_003_V010402_20240722T193843Z",
    engine="enmap"
)

Output

RasterioIOError: .../SPECTRAL_IMAGE.TIF: No such file or directory

Additional context

Update read_dataset_from_directory with glob and something like *{filename}.TIF? instead of {filename}.TIF.

[pont-us]

Thanks for the report! I've successfully reproduced the problem -- fix coming soon.

[pont-us]

@remoteJorge This is now fixed on the branch pont-3-directories. If you can install from this branch and confirm that the fix works for you I'll merge the changes and make a new release.

[remoteJorge]

Thanks @pont-us, it works!

Unrelated and optional suggestion: it could also be very useful to have the band center wavelengths stored in the dataset attributes, e.g. as enmap_dataset.attrs["wavelengths"] (or something similar, not sure what the best convention would be).

The metadata file already contains <wavelengthCenterOfBand> for every band. In principle, the wavelengths are always the same, but I think sometimes they can vary slightly depending on calibration, version, etc. It would be nice to have them included in the dataset attributes as a list, so users can be sure they are working with the exact values from the product.