Fix resampling of ancillary variables when also first class datasets

[mraspaud]

Sometimes, datasets from a scene are also an ancillary variable to another dataset. When resampling such a scene, only one of the instances of that dataset was actually replaced by the resampled version. This PR fixes this.

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

[codecov]

Codecov Report

Merging #652 into master will increase coverage by 0.04%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #652      +/-   ##
==========================================
+ Coverage   78.39%   78.44%   +0.04%     
==========================================
  Files         138      138              
  Lines       19980    20002      +22     
==========================================
+ Hits        15664    15690      +26     
+ Misses       4316     4312       -4

Impacted Files	Coverage Δ
satpy/scene.py	88.95% <100%> (+0.81%)	⬆️
satpy/tests/test_scene.py	99.46% <100%> (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 bc5ba34...27df564. Read the comment docs.

[coveralls]

Coverage increased (+0.04%) to 78.442% when pulling 27df564 on mraspaud:fix-resample-ancillary into bc5ba34 on pytroll:master.

[djhoese]

If I understand this as much or as little as I think I do, looks good to me. All this complication with ancillary variables makes me think we need to handle xarray.Dataset objects and that that may help simplify this...somehow.