Fix fill value for integer datasets, fix band assignment

[pnuu]

Since some recent changes in Satpy resampling a of mix of float and integer datasets (e.g. when using BackgroundCompositor) in to one output was broken in two ways in XArrayResamplerBilinear.

1. fill_value was set to np.nan even for the integer dataset
2. if one of these datasets had RGBA and another RGB bands, sometimes there were problems assigning the correct bands to the output dataset after resampling.

I'm not entirely sure why this happens, but this fixes it.

• Tests added
• Tests passed
• Passes git diff origin/master **/*py | flake8 --diff

[codecov]

Codecov Report

Merging #298 into master will increase coverage by 0.04%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #298      +/-   ##
==========================================
+ Coverage   92.60%   92.65%   +0.04%     
==========================================
  Files          43       43              
  Lines        9077     9104      +27     
==========================================
+ Hits         8406     8435      +29     
+ Misses        671      669       -2     

Impacted Files	Coverage Δ
pyresample/bilinear/xarr.py	94.84% <100.00%> (+0.86%)	⬆️
pyresample/test/test_bilinear.py	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update fc78bda...e17dd7b. Read the comment docs.

[mraspaud]

Thanks for the concise code and refactorings. Just the tests are missing.

[coveralls]

Coverage increased (+0.04%) to 92.652% when pulling e17dd7b on pnuu:bugfix_fill_value_and_bands into fc78bda on pytroll:master.

[pnuu]

I believe everything is now tested.