Add datasets_names attribute to cube models

[QRemy]

Add datasets_names attribute to each 3d model.
Adapt serialization so the models.yaml file contains datasets list for each model and remove the models and backgrounds lists from the datasets.yaml file.
If model.datasets_names is not set or set as "all" the model is applied to all datasets.

[codecov]

Codecov Report

Merging #2782 into master will increase coverage by 0.01%.
The diff coverage is 95.83%.

@@            Coverage Diff             @@
##           master    #2782      +/-   ##
==========================================
+ Coverage    92.3%   92.31%   +0.01%     
==========================================
  Files         144      144              
  Lines       16511    16521      +10     
==========================================
+ Hits        15240    15251      +11     
+ Misses       1271     1270       -1

Impacted Files	Coverage Δ
gammapy/spectrum/dataset.py	95.83% <100%> (+0.17%)	⬆️
gammapy/spectrum/flux_point.py	93.57% <100%> (+0.13%)	⬆️
gammapy/cube/fit.py	91.38% <100%> (+0.08%)	⬆️
gammapy/modeling/models/cube.py	91.9% <94.44%> (+0.12%)	⬆️
gammapy/utils/fits.py	95.16% <0%> (-1.51%)	⬇️
gammapy/maps/region.py	97.12% <0%> (+0.02%)	⬆️
gammapy/maps/regionnd.py	93.69% <0%> (+0.05%)	⬆️

---

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 14a6250...eb1c150. Read the comment docs.

[adonath]

Thanks @QRemy! I've left a few inline comments.

Additionally I think there is a number of places, where we create BackgroundModel objects internally (like MapDataset.create() or MapDatasetMaker.run()), can you make sure to set the correct dataset name to the BackgroundModel in these cases? And maybe add minimal regression tests to check the agreement of the names.

[adonath]

Thanks @QRemy, there is one remaining Travis fail in the analysis_mwl.ipynb notebook, that I'll fix in master.