Respect fmt: skip for compound statements on single line

[dylwil3]

Closes #11216

Essentially the approach is to implement Format for a new struct FormatClause which is just a clause header and its body. We then have the information we need to see whether there is a skip suppression comment on the last child in the body and it all fits on one line.

The ruff test fixture (credit: mostly created by Claude) contains several lines that seem to crash black, see:

• Crash with preview style fix_fmt_skip_in_one_liners for certain compound statements psf/black#4782
• Crash with preview style fix_fmt_skip_in_one_liners for function with annotated parameter and # fmt: skip on one line psf/black#4781

and another line containing a deviation that I think is unintentional for black, see:

• Preview style fix_fmt_skip_in_one_liners does not work for semi-colon separated statements in body of compound statement psf/black#4783

However, it's not clear whether we want the same behavior here? In this PR we have:

if True: print("this"); print("that") # fmt: skip

unchanged, but we do add a newline between the statements here (unlike Black):

print("this"); print("that") # fmt: skip

(see #11430 and #17331 .)

The other deviation from black is that we format the placement of the skip comment itself.

For review, it is simplest to go commit by commit.

[github-actions]

ruff-ecosystem results

Formatter (stable)

✅ ecosystem check detected no format changes.

Formatter (preview)

✅ ecosystem check detected no format changes.

[MichaReiser]

unchanged, but we do add a newline between the statements here (unlike Black):

print("this"); print("that") # fmt: skip

This seems unrelated to this PR as it is a pre-existing issue, right?

[MichaReiser]

This is great :) I suggest adding a few more tests with different comment placements. Just to make sure we handle them correctly.

[MichaReiser]

Can we derive the SuiteKind from the ClauseHeader to reduce the number of arguments?

[dylwil3]

I don't think so because I think we have to be told whether we're the last suite in the statement (but I think that's the only missing info)

[dylwil3]

The CI failure here https://github.com/astral-sh/ruff/actions/runs/18664070057/job/53211233008?pr=20633 was caused by this PR uncovering a pre-existing bug. See #21065 and #21067 . Will rebase once that PR is merged and then continue working on this!

[dylwil3]

Cut down a comment-related hydra head and more grew in its place. This snippet causes a panic because the indented comment is not marked as formatted. Annoyingly it is attached to bar() (because there is no orelse node).

for x in it: foo()
    # comment
else: bar() # fmt: skip

Converting to draft while I sort it out 😄