Add --add-ignore for adding ruff:ignore comments

[ntBre]

Summary

This PR adds the --add-ignore flag as an alias to --add-noqa. In preview, this adds
ruff:ignore comments with human-readable names instead of noqa comments with codes.

I initially drafted this PR assuming that --add-ignore would be a separate flag that only
functioned in preview mode, while --add-noqa would be available in both preview and stable, but
after talking with Micha realized that we probably want to go ahead and push preview users toward
--add-ignore entirely rather than giving them the choice. I think the one downside of the alias
approach is that --add-noqa still exists in preview but will still add ruff:ignore comments.

In cases where an existing noqa or ruff:ignore comment is present, I opted to preserve the
existing comment and to append the new comment. So something like this:

import math  # noqa: RUF100

becomes:

import math  # noqa: RUF100 # ruff:ignore[unused-import]

instead of also converting the existing noqa comment. I think we can defer to one of our planned
lint rules for the conversion (and joining) instead.

As also shown in this example, the flag adds human-readable names rather than rule codes and prefers
the same placement as a noqa comment, at the end of the line. The one exception to this is an
existing ruff:ignore comment on its own line. In that case, we will reuse and extend the existing
comment instead of adding an additional trailing comment, which I think is a pretty nice balance.

This PR also fixes #26287, which I discovered while working on this and turned out to be easy to fix
while I was here.

Test Plan

Many new CLI tests exercising the various cases Codex and I could come up with. As the server test
shows, this actually applies to the LSP as well, where the Disable for this line action now also
inserts a ruff:ignore comment with a rule name instead of a noqa comment.

[astral-sh-bot]

ruff-ecosystem results

Linter (stable)

✅ ecosystem check detected no linter changes.

Linter (preview)

✅ ecosystem check detected no linter changes.

[ntBre]

As part of this PR, I'm checking how well --add-noqa works on our ecosystem projects, first running check --select ALL to collect a baseline, then adding noqa comments, and then checking again. One immediate thing I notice is a huge number of added I001 diagnostics (5278 on all ecosystem projects). These aren't particularly alarming, as they seem to be a result of import lines becoming too long with the new comment.

Then there are 923 new RUF100 diagnostics from a variety of rules that seem not to be suppressible. One example is RUF101: https://play.ruff.rs/45d0e92f-fe87-437e-af87-9a83b099ff5b. Another is PLR2044: https://play.ruff.rs/9bcf9da7-4a50-494b-b6d0-cda53188b876. Another is SIM109: https://play.ruff.rs/922b0cf5-b97e-4c9f-8bdc-e300283f35fe.

We can also introduce syntax errors, mostly via UP031 from what I've seen so far:

❯ cat try.py
print("A long string \  # noqa: UP031
   with a continuation on line %d" % 2)

❯ python try.py
  File "try.py", line 1
    print("A long string \  # noqa: UP031
          ^
SyntaxError: unterminated string literal (detected at line 1)

There are only 10 of these, but we can also disrupt (our detection of?) script metadata and trigger INP001:

# /// script  # noqa: CPY001
# requires-python = ">=3.11"
# dependencies = [
#     "uv==0.11.13",

I'll plan to open a separate issue or issues for these, but for now I'll just try to make sure --add-ignore doesn't make the situation any worse, at least excluding the isort rules, since the longer ruff:ignore[rule-name] will obviously exacerbate that.