Improve performance of loading range suppressions

[amyreese]

WIP, not fully baked, checkpointing progress:

• creates new block range indexer to track indent/dedent ranges
• pass indexer to suppression loader
• new implementation of loading from tokens based on comment ranges and block range index
• doesn't need to walk tokens forward/backward looking for matching indent scopes
• has issue with first comment in block preceding the indent token, and block range index not knowing

issue #22087

[astral-sh-bot]

ruff-ecosystem results

Linter (stable)

✅ ecosystem check detected no linter changes.

Linter (preview)

✅ ecosystem check detected no linter changes.

Formatter (stable)

✅ ecosystem check detected no format changes.

Formatter (preview)

✅ ecosystem check detected no format changes.

[codspeed-hq]

CodSpeed Performance Report

Merging this PR will degrade performance by 17.64%

_{Comparing amy/suppression-perf (c7b528e) with main (f97da18)}

Summary

❌ 5 regressed benchmarks
✅ 48 untouched benchmarks

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

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	linter/all-with-preview-rules[numpy/globals.py]	831.7 µs	914.3 µs	-9.04%
❌	Simulation	linter/all-with-preview-rules[unicode/pypinyin.py]	2.2 ms	2.7 ms	-17.64%
❌	Simulation	linter/all-with-preview-rules[numpy/ctypeslib.py]	5.2 ms	5.7 ms	-8.8%
❌	Simulation	linter/all-with-preview-rules[large/dataset.py]	22.3 ms	23.5 ms	-5.14%
❌	Simulation	linter/all-with-preview-rules[pydantic/types.py]	10.2 ms	11 ms	-7.48%

[amyreese]

codspeed not happy 🤪

[MichaReiser]

codspeed not happy 🤪

dbg statements are performance killers

[MichaReiser]

I prefer avoiding adding new state to Indexer because it's state that we can't release until the file has finished checking. That's why we generally prefer re-computing information unless doing so is expensive or has to be repeated in many places (recomputing can often also be cheaper).

Are there cases where the implementation in #21441 (review) falls short, making computing this information ahead of time necessary?