Fix numpy 1.24 warnings

[ianthomas23]

Fixes #1216.

Summary of fixes:

• Conversion of float array containing NaN to integer:

  • Simple uses now have a call to np.nan_to_number to convert np.nan to 0, which is the number that has always been returned in this scenario.
  • The more complicated code in category_binning already determines a mask of the NaN values, so those NaN values are set to 0 (they could be any valid finite number) before the conversion to int.
  • In Canvas.raster we avoid the warning by using a fill_value of 0 for integer dtypes as that is what the docstring says is used in this scenario.

• Warnings about finding max or min of an "all-nan slice" are specifically ignored as they already return the expected value of nan in this situation.

[ianthomas23]

This is consistent with what is in test_dask.py.

[codecov]

Codecov Report

Merging #1218 (26c8f4a) into main (8092f4d) will increase coverage by 0.32%.
The diff coverage is 95.35%.

@@            Coverage Diff             @@
##             main    #1218      +/-   ##
==========================================
+ Coverage   84.52%   84.85%   +0.32%     
==========================================
  Files          35       35              
  Lines        8369     8570     +201     
==========================================
+ Hits         7074     7272     +198     
- Misses       1295     1298       +3     

Impacted Files	Coverage Δ
datashader/data_libraries/pandas.py	100.00% <ø> (ø)
datashader/reductions.py	84.70% <93.24%> (+1.59%)	⬆️
datashader/compiler.py	91.07% <100.00%> (+0.50%)	⬆️
datashader/core.py	88.28% <100.00%> (+0.04%)	⬆️
datashader/data_libraries/dask.py	95.23% <100.00%> (+0.07%)	⬆️
datashader/data_libraries/dask_xarray.py	98.95% <100.00%> (ø)
datashader/transfer_functions/__init__.py	87.03% <100.00%> (+0.12%)	⬆️
datashader/utils.py	81.86% <100.00%> (+2.61%)	⬆️

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more