perf(linter): performance improvement for css semantic model

[togami2864]

Summary

According to the profile, get_rule_by_range and process_selector significantly affect performance.

I implemented the following improvements to address these issues:

• Replaced FxHashMap with BTreeMap to search range_to_rule more efficiently.
• Used Cow to minimize unnecessary memory allocation in process_selector
• Removed unused functions.

Test Plan

CI should pass

[codspeed-hq]

CodSpeed Performance Report

Merging #4044 will degrade performances by 6.6%

_{Comparing togami2864:perf/semantic-css (341558f) with main (b3724b2)}

Summary

⚡ 5 improvements
❌ 1 regressions
✅ 101 untouched benchmarks

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

Benchmarks breakdown

	Benchmark	main	togami2864:perf/semantic-css	Change
⚡	css_analyzer[bootstrap_18416142857265205439.css]	255.9 ms	203.6 ms	+25.66%
⚡	css_analyzer[bulma_5641719244145477318.css]	784.5 ms	645.5 ms	+21.53%
⚡	css_analyzer[foundation_11602414662825430680.css]	158.4 ms	139.9 ms	+13.25%
⚡	css_analyzer[tachyons_11778168428173736564.css]	151.7 ms	112.7 ms	+34.53%
❌	react.production.min_3378072959512366797.js[cached]	1.8 ms	2 ms	-6.6%
⚡	db_17847247775464589309.json[uncached]	16.3 ms	15.2 ms	+7.33%

[ematipico]

Awesome job @togami2864 !!