Add modes for nan and inf handling to adaptive algo

[svank]

I found myself reprojecting images with NaNs and infs, so I added a few modes for how to deal with them. The default remains the previous behavior, which just propagates these values with no special handling.

The constant mode isn't strictly necessary, as one could use np.nan_to_num on the input array before reprojecting, but implementing this in reproject avoids having to copy the array---maybe that's important for dask use cases.

[codecov]

Codecov Report

Merging #380 (a4312de) into main (661963d) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##             main     #380   +/-   ##
=======================================
  Coverage   93.44%   93.44%           
=======================================
  Files          25       25           
  Lines         870      870           
=======================================
  Hits          813      813           
  Misses         57       57           

Files Changed	Coverage Δ
reproject/adaptive/core.py	95.91% <ø> (ø)
reproject/adaptive/high_level.py	100.00% <ø> (ø)

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

[astrofrog]

I've rebased this and have also changed bad_val_mode to bad_value_mode - I try and avoid abbreviations if it only saves a couple of letters. Hope that's ok! Will merge once CI passes.

[svank]

@astrofrog Thanks for tying off these PRs! Happy to see another release 🎉