Table unicode sandwich follow-up: allow Table FITS reader to return bytestrings #6122

Currently in Py3 one gets unicode character arrays when reading string data from a FITS file. This translation is done in the the machinery within io.fits that lazily auto-converts any string column to U-type for cols that can be decoded to ASCII. See also here.

After #5700 there is the opportunity to have Table.read('file.fits') return a Table with 'S'-type columns, thus saving a factor of 4 in memory.

However, without implementing the unicode sandwich directly in io.fits (as @embray has prototyped), we have to be careful that direct use of HDU columns still returns unicode in Py3 while when used for reading a Table it would return bytes. It isn't immediately obvious to me how to implement this dual-behavior (in terms of API), but I haven't looked too hard.

There is also the question of whether this should be the default or requires opt-ing in.

cc: @saimn