Warnings from new `unit_parse_strict='warn'` settings

Description

#11843 introduced an improved validation setting for non-standard units in FITS headers and made the 'warn' level the default (contrary to the docstring description). This issues new UnitsWarnings in a number of circumstances like the RTD builds in #12784.

Expected behavior

The unit_parse_strict default should be documented in the read_table_fits docstring.

The specific case of the RTD warnings from timeseries is probably best addressed by explicitly setting Table.read(hdu, format='fits', unit_parse_strict='silent') in kepler_fits_reader, since that function is performing its own custom treatment of nonstandard units (or should this make better use of the u.add_enabled_units mechanism as well?)

But we probably should expect a number of similar warnings, possibly treated as errors, to pop up downstream from this change.

System Details

5.1.devx

[dhomeier]

No, the errors just don’t appear in the 5.0.1 backport since #11843 is on the 5.1 branch. Also it is not just a docs issue since the warnings will start popping up in anything that uses read_table_fits or Table.read directly, so it should probably be dealt with before the 5.1 release.

[saimn]

There is still the case of BJD - 2454833 which is not handled here. And even with the “generic” parser, e or pixels would not be valid units. So if those files contain invalid units, I guess that’s up to timeseries’ maintainers to choose to use unit_parse_strict='silent' or not (would make sense since it was the previous behavior), but I don’t think that’s a fits or units issue 😃.