Fix gm nodata

[emmaai]

I changed my mind on where to fix gm nodata issue, based on the following reasons:

• Both s2 and ls inherit the same plugin and hence same function in computing the gm. They can have their different ways in defining nodata for extra bands other than the input bands.
• The nodata is defined in production definition yaml. All datasets should respect and follow the same definition such that dc.load and rioxarray.open will result the same nodata information. For future improvement ref issue #131
• The same "hardcode" approach also caused discrepancy between data type of GeoTIFF and one in product def, e.g., count being uint16 in GeoTIFF while int16 in product def, which results in int16 in dc.load while uint16 with other tools
• gdal doesn't fully respect default nodata=nan for float when generating overviews either. Thus it'd be good that we honour nodata definition every level. In this PR, I allowed nodata not defined for count band because of a) data type discrepancy mentioned above, b) nodata not presents in raster hence no trouble atm

I decide to address the coding issues later while keep the path clear for now.

[codecov[bot]]

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Project coverage is 78.29%. Comparing base (0aa9773) to head (9bba420). Report is 2 commits behind head on develop.

Additional details and impacted files

```diff @@ Coverage Diff @@ ## develop #135 +/- ## =========================================== + Coverage 78.24% 78.29% +0.04% =========================================== Files 38 38 Lines 3517 3525 +8 =========================================== + Hits 2752 2760 +8 Misses 765 765 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.