Avoid treating dataclasses.KW_ONLY as typing-only

[charliermarsh]

Summary

Closes #12859.

[codspeed-hq]

CodSpeed Performance Report

Merging #12863 will degrade performances by 4.1%

_{Comparing charlie/kw-data (d051027) with main (899a523)}

Summary

⚡ 1 improvements
❌ 1 regressions
✅ 30 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	main	charlie/kw-data	Change
⚡	linter/all-rules[numpy/globals.py]	765 µs	723.1 µs	+5.8%
❌	linter/default-rules[pydantic/types.py]	1.8 ms	1.9 ms	-4.1%

[github-actions]

ruff-ecosystem results

Linter (stable)

✅ ecosystem check detected no linter changes.

Linter (preview)

✅ ecosystem check detected no linter changes.

[AlexWaygood]

These special cases hurt so badly. But it's not our fault, just something we have to live with since Python's semantics are what they are!

PEP 649 can't land soon enough :(