Fix crash on fmt: skip after the last base of a skipped class header - #5633

Description

A class header with # fmt: skip crashes when its last base or keyword argument also has a # fmt: skip:

class A(
    B,  # fmt: skip
):  # fmt: skip
    pass

$ black repro.py
error: INTERNAL ERROR: Black 9a1f64e on Python (CPython) 3.12.15 produced invalid code:
cannot parse: repro.py:3:4
        pass
       ^
ParseError: bad input

The output it can't parse has the header's ): inside the comment:

class A(
    B,  # fmt: skip):  # fmt: skip
    pass

The same happens in these cases:

• B without the comma, or more than one base with the # fmt: skip on the last one;
• metaclass=M;
• a decorated class, or a nested class.

26.10.1 crashes the same way. def, if, while, for and with headers with the same two comments format fine.

What happens:

• The inner # fmt: skip hides B,. It takes its comment, together with the newline after it, from the prefix of the header's ).
• The header's # fmt: skip then turns the whole header into one standalone comment, and the hidden B, # fmt: skip needs that newline back.
• For def and the other statements, the brackets sit inside a node (parameters, atom), so the hidden part goes through stringify_node. That adds a newline after a standalone comment whose last line has a # (Fix crash on a fmt: skip header containing an inner skipped bracket #5597).
• A class's parentheses are direct children of classdef, so the hidden base went through stringify_standalone_comment instead. That only added the newline when the next leaf's prefix had one, and here the prefix had lost it.

stringify_standalone_comment now also adds the newline when the last line has a #, as stringify_node does. The header is then kept exactly as written.

To check that nothing else changes:

• I generated 1,079 inputs and formatted each in stable and preview mode (2,158 runs). They are compound statement headers (class, def, if, while, for, with, except) and statements, with # fmt: skip, # fmt:skip, # type: ignore or another comment on one line, and # fmt: skip on two or three lines.
  • Main crashes on 56 of the runs and this branch on 10.
  • Every run that main formats gives the same output on this branch.
  • The 10 that still crash are a decorator ending in ) # fmt: skip right before class A( # fmt: skip. That is unstable formatting ("produced different code on the second pass") on main and on 26.10.1 too, and this change doesn't fix it.
• The 12 files in a Python 3.12 install (stdlib and site-packages) that contain fmt: skip or fmt: off format to identical output on main and on this branch.

Checklist - did you ...

• Implement any code style changes under the --preview style, following the stability policy? (A crash fix, so stable as the policy allows; code that formatted before formats the same.)
• Add an entry in CHANGES.md if necessary?
• Add / update tests if necessary? (New case fmtskip_class_header_base.py: the cases above and a nested class with a docstring. It fails on main with the crash.)
• Add new / update outdated documentation? (Not needed.)

Test results:

• pytest tests: 725 passed. The only failures are test_expression_diff_with_color and test_piping_diff_with_color, which fail the same way on main in my terminal (no ANSI codes in the output).
• pytest tests --run-optional jupyter -m jupyter: 89 passed.
• black --check . (run_self) and the pre-commit hooks (isort, flake8, mypy, prettier, end-of-file, trailing whitespace) pass.