FITS header sanity

[teuben]

Various items in the cube FITS header need to be checked with 3rd party packages to see if we can switch between VLSR and FREQ for example. The latest version (10-jan-2021) now reports the history of how the dataset was created.
There are still some comments in the fits comment field where we need to pay attention (e.g. SPECSYS, RADECSYS). The issue around how the VLSR is supposed to be written now seems to be ZSOURCE, not VLSR or VSOURCE.

The only non-standard keyword we write is OBSNUM (even for a combination, which is arguably wrong)

[teuben]

As an example, for the benchmark IRC data, in the fits file we have 350.303 km/s at the reference pixel, but once in CASA, this becomes 349.894. cdelt3 is 1.015921 km/s, but in CASA it's 1.013554. Perhaps we need to supply ALTRVAL/ALTRPIX?

[teuben]

It seems CASA is using VRAD, not VELO-LSR, on export. The circle with miriad is not rounded yet though.

[teuben]

aplpy still has an issue with a cube. claims not a proper wcs.