bug#59477: python-tests fail without tree-sitter

[Mattias Engdegård]

> But if I run this in a buffer manually I get the ending newline. I’m not sure 
> what’s the cause of that. Bisecting give 
> 7c5d4348330b206aff1f8e5bc4fd241d6a6dc0b5, but that commit doesn’t change 
> anything filling-related. 

No idea really, but it might have something to do with the fact that the 
changes move the assignments

  (setq-local font-lock-defaults
              `(,python-font-lock-keywords
                nil nil nil nil
                (font-lock-syntactic-face-function
                 . python-font-lock-syntactic-face-function)))
  (setq-local syntax-propertize-function
              python-syntax-propertize-function)

so that they are executed after

  (when python-indent-guess-indent-offset
    (python-indent-guess-indent-offset))

instead of before. `python-indent-guess-indent-offset` has the side-effect of 
setting syntax properties, in particular for the string terminator 
(triple-quote in the test).

This is important, because python-fill-string (called as part of fill-paragraph 
in the test) assumes this having already been done and if not, str-end-pos 
isn't computed correctly and things take a turn for the worse after that.

Stefan probably knows better how this is supposed to work, but presumably 
python-fill-string should take measures to ensure accurate syntax properties 
before doing things like

  (re-search-forward (rx (syntax string-delimiter)) nil t)

and so on. Sorry about not being of much help here.