Refactor I/O

[sfinkens]

Move masking/correction logic from gac_io to the readers so that the satpy reader (which is using pygac as a library) returns the same data as pygac-run writes to hdf5.

The refactoring doesn't change the output expect for one case:

• The current version doesn't catch NaN lat/lon so that they end up as -2147483648 (the minimum integer) in the hdf5 file. That will be fixed by this PR.

Other features added:

• Make coordinate interpolation optional
• Make clock drift adjustment optional
• Add TLE dir & name attributes to the reader (-> no PYGAC_CONFIG_FILE required for use in satpy)

---

The changes have been validated with 130 "best of corruption" orbits. Summary:

• Out of 129 orbits 7 differ in the refactored version. The other 122 are binary identical.
• If two files differ, it's only the mask, not the data. If there are two valid data points at a given pixel, they are identical.
• The differences are all caused by very few NaNs in the lat/lon arrays. Here the masking of the new version is clearly better.
• Broken-scan-motor orbits are identical, too. The differences I reported before were due to a unit/channel mismatch. This is now fixed.

You can find some plots here: https://public.cmsaf.dwd.de/data/sfinkens/pygac_refactoring/ . Display them in 100%. You might have to clean your screen in order to see the mask difference in some cases ;)

TODO:

• Thorough validation of the altered results
• Fix tests
• Add more tests

[mraspaud]

Thanks for the comprehensive refactor and adding all the tests! Just a few small points to fix.