Default floating point bands to nodata=nan when writing COGs.

[SpacemanPaul]

Reason for this pull request

Issue #1595

Proposed changes

• Always write nodata=nan to COGs for all floating point bands (unless another nodata value is explicitly set) and test.

• Update tests to pass with numpy 2.x and numpy 1.x

• Pin moto<5 (Moto API has changed slightly in v5 and several tests will need to be rewritten to support).

• Closes Important fix proposal: Implement setting "nan" nodata value for floating point rasters with no nodata attribute  #1595

• Tests added / passed

• Fully documented, including docs/about/whats_new.rst for all changes

---

📚 Documentation preview 📚: https://datacube-core--1602.org.readthedocs.build/en/1602/

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 91.78%. Comparing base (8b46c22) to head (783124f).

Additional details and impacted files

@@           Coverage Diff            @@
##           develop    #1602   +/-   ##
========================================
  Coverage    91.78%   91.78%           
========================================
  Files          132      132           
  Lines        14637    14640    +3     
========================================
+ Hits         13434    13437    +3     
  Misses        1203     1203           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[robbibt]

I think this is working nicely, thanks @SpacemanPaul! If I try and write out a floating point array with no nodata attribute, I now get nodata=nan in the COG:

import datacube
from datacube.utils.cog import write_cog
dc = datacube.Datacube()

ds = dc.load(
    product="ga_ls8c_ard_3",
    measurements=["nbart_red"],
    time=("2019-01-10", "2019-01-15"),
    x=(149.036528545, 149.233835134),
    y=(-35.198132594, -35.384897602), 
    group_by="solar_day",
)

# Convert to float
da_float = ds.nbart_red / 10000
assert "nodata" not in da_float.attrs

# Export to file
write_cog(da_float, "test_missingnodata.tif", overwrite=True)
!gdalinfo test_missingnodata.tif

I also get nodata=nan in the COG if it is explicitly declared on the Xarray too:

import numpy as np
da_float.attrs["nodata"] = np.nan

# Export to file
write_cog(da_float, "test_nodatanan.tif", overwrite=True)
!gdalinfo test_nodatanan.tif

And custom nodata is still preserved if provided on the Xarray:

da_float.attrs["nodata"] = -999

# Export to file
write_cog(da_float, "test_nodatacustom.tif", overwrite=True)
!gdalinfo test_nodatacustom.tif