I made it to agree as much as possible with the rules from CLDR (see:
https://github.com/unicode-org/cldr/blob/main/common/collation/th.xml).

It seems to be impossible to follow the CLDR rules

  &[before 1]๚<ฯ # should be "variable"

and

  &๛<ๆ # should be "variable"

exactly though. These ask for a primary difference in punctuation
characters whose primary weight should be "IGNORE". But using a
secondary differnence instead still sorts the test data correctly and
the previously used collation in th_TH used tertiary differences for
these characters.

There was old localedata/th_TH.in test data in TIS-620 encoding which
was not used (it was not in the localedata/Makefile). I converted this
to UTF-8 and moved it to localedata/th_TH.UTF-8.in and added it to
localedata/Makefile.

Using the existing collation rules in the th_TH locale did not sort that
test file completely correct, I think my new collation rules based on
iso14651_t1 are better.