GlobBiomass: several bug fixes

[adamjstewart]

This PR fixes the following bugs:

• GlobBiomass data is from 2010, although it isn't in the filename
• Mask is for pixelwise regression, use float32, not long
• Length of dataset was doubled due to unspecific filename_glob

I also improved the documentation to include definitions and units of AGB and GSV.

I didn't do these in this PR, but some suggestions for future contributions to this dataset:

• Add download support, URLs are public
• Add support for file extensions other than .tif
• Split estimation mask and uncertainty mask into two different dict keys (backwards incompatible)
• Treat measurements like bands and allow return both
• Treat measurements AND errors like bands and allow return both
• Add data module, our first real pixelwise regression dataset