Fix formatting oscillation in if-then-else branches

[MisterDA]

When a comment sits between a keyword (then/else) and certain branch
expressions (begin...end wrapping match/try/function/ifthenelse, or bare
match/try/function/ifthenelse), the comment would alternate between
"after keyword" and "before expression" placements across formatting
iterations. This caused branch_pro to compute different indentation each
time, producing "formatting did not stabilize after 10 iterations".

The fix detects these oscillation-prone patterns and ensures stable
formatting regardless of the initial comment placement:

• In Fmt_ast, when a branch matches needs_raw_cmts_after_kw, extract
  "after keyword" comments without breaks and override branch_pro using
  Params.raw_cmts_branch_pro (which computes mode-appropriate
  break+indent+comment output).

• In Params, branch_pro_with_cmts detects comments "before expression"
  (the alternate placement) and forces a break, producing the same layout
  as the "after keyword" path.

• For begin...end branches, branch_pro checks for comments before the
  inner expression (not just the begin...end node) to handle the case
  where the comment migrates inside the begin...end scope.

Affected modes: Fit_or_vertical, Compact (profile=conventional).

Closes #2799.

[MisterDA]

Thanks for the review, here's a revised version of the patch with your comments.
Not sure where the build failures on Windows come from, and I think they're unrelated.

[Julow]

I really don't like this cmts_after_kw_raw argument. I tried to get rid of it by passing Cmts.fmt_after ~pro:noop ~epi:noop .. to ~cmts_after_kw but it's very hard due to the many different states this creates.

I won't merge a new argument to get_if_then_else.

[MisterDA]

Claude found another workaround. Let me know if it's satisfying or not.
EDIT: not sure how to reproduce the CI failures.