Add UVFITS support for large antenna numbers

[kettenis]

Follow the scheme documented in AIPS memo 117 (revised, 19 June 2017)
and use separate SUBARRAY, ANTENNA1 and ANTENNA2 random parameters
when antenna numbers larger than 255 are encountered when writing
UVFITS files.

Add support for reading such UVFITS files back in as well.

[kettenis]

With these changes, the OSKAR MS with 512 antennas provided to me by Robert Braun can be exported and can be loaded into AIPS using the FITLD task. The resulting data structures look fine when explored using ParselTongue, but some AIPS tasks are unhappy. I suspect that this is a problem with the characteristics of the data produced by OSKAR. A VLBI data set that has been written in the new format (by forcing the codepath that writes out SUBARRAY, ANTENNA1 and ANTENNA2 instead of BASELINE random parameters) seems to behave just fine.

Converting the OSKAR data set back into MS seems to work as well, with some expected loss of metadata.

The CASA regression tests still pass with this change.

[tammojan]

I guess there's no existing unit test that could be used to check the new behavior?

The changes overall look good to me.

[kettenis]

Pushed a new version with the requested changes. Also adjusted the commit message to match the title change done by @aroffringa