perf: Reduce best fitting allocations

[MichaReiser]

Let's try this again, now that we have more best fitting elements.

Summary

This PR removes the number of allocations necessary when using BestFitting.

Best fitting performs the following allocations today:

1. It allocates one Vec in which it stores all variants
2. For each variant, allocate a Vec<FormatElement> that stores the variant's IR elements

This PR eliminates 2. It uses StartBestFittingEntry and EndBestFittingEntry markers to identify the start/end of a variant. The outer vector is still required because it otherwise gets fairly complicated to identify the start/end of a variant in case we encounter nested best fittings.

This won't have as much impact anymore when #7475 lands but we'll need best fitting to implement Black's preview string processing

Test Plan

cargo test

[codspeed-hq]

CodSpeed Performance Report

Merging #7411 will improve performances by 7.76%

_{Comparing best-fitting-reduce-allocations (441a75f) with main (2421805)}

Summary

⚡ 4 improvements
✅ 21 untouched benchmarks

Benchmarks breakdown

	Benchmark	main	best-fitting-reduce-allocations	Change
⚡	formatter[unicode/pypinyin.py]	3.8 ms	3.6 ms	+3.21%
⚡	formatter[large/dataset.py]	61.7 ms	57.3 ms	+7.76%
⚡	formatter[pydantic/types.py]	21.4 ms	20.8 ms	+3.18%
⚡	formatter[numpy/ctypeslib.py]	11.1 ms	10.7 ms	+3.64%

[MichaReiser]

Current dependencies on/for this PR:

• main
  • PR perf: Reduce best fitting allocations #7411 👈
    • PR perf(formatter): Allocation free interned #7443

This comment was auto-generated by Graphite.