Add NDData support as data factories to glue

[kakirastern]

Description

Add astropy.nddata support as data factories in order to offer an alternative means to load more specific data types, such as FITS files, but using the NDData, making the manipulation of units, mask, uncertainty, and meta of any FITS data more manageable in the context of glue.

[codecov]

Codecov Report

Merging #2164 into master will decrease coverage by 0.00%.
The diff coverage is 84.84%.

@@            Coverage Diff             @@
##           master    #2164      +/-   ##
==========================================
- Coverage   87.87%   87.87%   -0.01%     
==========================================
  Files         246      247       +1     
  Lines       22724    22788      +64     
==========================================
+ Hits        19969    20024      +55     
- Misses       2755     2764       +9     

Impacted Files	Coverage Δ
glue/core/data_factories/nddata.py	77.14% <77.14%> (ø)
glue/core/data.py	90.71% <93.33%> (+0.07%)	⬆️
glue/core/data_factories/__init__.py	100.00% <100.00%> (ø)
glue/__init__.py	70.37% <0.00%> (+3.70%)	⬆️

---

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 cf93252...0b089d7. Read the comment docs.

[astrofrog]

Thanks for the PR! This introduces a very big change which is to add mask and uncertainty to the main data object, which I don't think we want to do. In particular, each attribute could in principle have an uncertainty. I think a better approach would be for now to store the mask and uncertainty in attributes with special names which then doesn't require any changes to the main data class.

This also overlaps discussion wise with glue-viz/glue-astronomy#17 - which is about translating NDData <-> Data and once that is complete we can leverage that to simplify the FITS as NDData reader. However, I think that to make things easier to manage, this data factory here should also be added to glue-astronomy, not glue-core (to keep all the NDData stuff in a single place)

[astrofrog]

More generally, I wonder if we should just be doing NDData.read here and making that work properly for FITS in astropy. Alternatively, glue-solar could implement factories that delegate to <some NDData subclass>.read

[kakirastern]

Sure, let me make the changes soon