Tweak breaking groups for comprehensions

[charliermarsh]

Summary

Fixes some comprehension formatting by avoiding creating the group for the comprehension itself (so that if it breaks, all parts break on their own lines, e.g. the for and the if clauses).

Closes #6063.

Test Plan

Bunch of new fixtures.

[github-actions]

PR Check Results

Benchmark

Linux

group                                      main                                   pr
-----                                      ----                                   --
formatter/large/dataset.py                 1.00      9.1±0.51ms     4.5 MB/sec    1.04      9.5±0.37ms     4.3 MB/sec
formatter/numpy/ctypeslib.py               1.01  1846.3±85.62µs     9.0 MB/sec    1.00  1833.3±70.34µs     9.1 MB/sec
formatter/numpy/globals.py                 1.00   207.2±11.54µs    14.2 MB/sec    1.00   207.5±11.04µs    14.2 MB/sec
formatter/pydantic/types.py                1.00      3.8±0.12ms     6.8 MB/sec    1.05      3.9±0.15ms     6.5 MB/sec
linter/all-rules/large/dataset.py          1.04     13.1±0.46ms     3.1 MB/sec    1.00     12.6±0.39ms     3.2 MB/sec
linter/all-rules/numpy/ctypeslib.py        1.03      3.3±0.14ms     5.1 MB/sec    1.00      3.2±0.08ms     5.2 MB/sec
linter/all-rules/numpy/globals.py          1.01   464.5±20.72µs     6.4 MB/sec    1.00   458.6±16.25µs     6.4 MB/sec
linter/all-rules/pydantic/types.py         1.05      6.1±0.26ms     4.2 MB/sec    1.00      5.8±0.18ms     4.4 MB/sec
linter/default-rules/large/dataset.py      1.08      6.5±0.17ms     6.3 MB/sec    1.00      6.0±0.15ms     6.8 MB/sec
linter/default-rules/numpy/ctypeslib.py    1.07  1391.0±70.38µs    12.0 MB/sec    1.00  1296.0±32.04µs    12.8 MB/sec
linter/default-rules/numpy/globals.py      1.04    166.8±7.77µs    17.7 MB/sec    1.00    160.6±6.51µs    18.4 MB/sec
linter/default-rules/pydantic/types.py     1.05      2.9±0.10ms     8.9 MB/sec    1.00      2.7±0.11ms     9.3 MB/sec

Windows

group                                      main                                    pr
-----                                      ----                                    --
formatter/large/dataset.py                 1.00     12.5±0.46ms     3.3 MB/sec     1.00     12.5±0.68ms     3.3 MB/sec
formatter/numpy/ctypeslib.py               1.02      2.4±0.14ms     6.9 MB/sec     1.00      2.4±0.11ms     7.0 MB/sec
formatter/numpy/globals.py                 1.00   261.1±18.55µs    11.3 MB/sec     1.05   275.0±21.35µs    10.7 MB/sec
formatter/pydantic/types.py                1.00      5.2±0.26ms     4.9 MB/sec     1.02      5.3±0.26ms     4.8 MB/sec
linter/all-rules/large/dataset.py          1.00     17.1±0.65ms     2.4 MB/sec     1.02     17.4±0.60ms     2.3 MB/sec
linter/all-rules/numpy/ctypeslib.py        1.00      4.4±0.25ms     3.8 MB/sec     1.02      4.5±0.20ms     3.7 MB/sec
linter/all-rules/numpy/globals.py          1.02   542.8±33.74µs     5.4 MB/sec     1.00   532.7±29.03µs     5.5 MB/sec
linter/all-rules/pydantic/types.py         1.00      7.6±0.34ms     3.3 MB/sec     1.02      7.8±0.29ms     3.3 MB/sec
linter/default-rules/large/dataset.py      1.00      8.6±0.23ms     4.8 MB/sec     1.00      8.6±0.27ms     4.7 MB/sec
linter/default-rules/numpy/ctypeslib.py    1.01  1792.3±125.82µs     9.3 MB/sec    1.00  1781.2±86.13µs     9.3 MB/sec
linter/default-rules/numpy/globals.py      1.00   204.3±11.58µs    14.4 MB/sec     1.01   205.4±14.74µs    14.4 MB/sec
linter/default-rules/pydantic/types.py     1.00      3.9±0.17ms     6.6 MB/sec     1.00      3.9±0.16ms     6.6 MB/sec

[MichaReiser]

Please update the PR summary with the similarity index before/after your change to ensure we're improving the formatting overall, and aren't over-fitting on a single usage.

It may also be good to diff the Black/Ruff differences between the new and old ruff version, e.g. using the django project and review the changes to ensure we aren't regressing anything.

[konstin]

stability index (caveat: i picked the lastest run from main instead of the branch base):

name: PR, main
zulip: 0.992 0.992
django: 0.992 0.991
warehouse: 0.978 0.978
transformers: 0.993 0.993
cpython: 0.759 0.759
typeshed: 0.723 0.723

I agree with manually checking django though, i found this helpful in the past.

[konstin]

surprised there are no black fixtures for this

[charliermarsh]

(Reviewed manually, happy with the changes.)