Enable preview mode during benchmarks

[zanieb]

Split out of #7195 so benchmark changes from enabling additional rules can be reviewed separately.

[codspeed-hq]

CodSpeed Performance Report

Merging #7208 will degrade performances by 5.6%

_{Comparing zanie/preview-benchmark (32e5cab) with main (6566d00)}

Summary

❌ 5 (👁 5) regressions
✅ 20 untouched benchmarks

Benchmarks breakdown

	Benchmark	main	zanie/preview-benchmark	Change
👁	linter/all-rules[numpy/globals.py]	4 ms	4.1 ms	-2.58%
👁	linter/all-rules[numpy/ctypeslib.py]	33.2 ms	34.5 ms	-3.93%
👁	linter/all-rules[unicode/pypinyin.py]	15.4 ms	15.7 ms	-2.29%
👁	linter/all-rules[pydantic/types.py]	69.1 ms	72.4 ms	-4.59%
👁	linter/all-rules[large/dataset.py]	154.5 ms	163.7 ms	-5.6%

[github-actions]

PR Check Results

Ecosystem

✅ ecosystem check detected no changes.

[zanieb]

Another option is to add new benchmarks for preview mode. Probably a good way to separate performance changes in preview and stable?

[MichaReiser]

Another option is to add new benchmarks for preview mode. Probably a good way to separate performance changes in preview and stable?

Sounds good to me. Separating preview and stable benchmarks can help narrow down a regression or catch unintentional performance regressions when supposedly only changing preview code. It could then be taken off lightly as: this is preview only when in fact, we just regressed stable.

[zanieb]

I think I'd like to include this as a general improvement and we can look for a contributor for splitting them into separate benchmarks?