fuzzy_nucleo: Optimize path matching with CharBag prefilter and add benchmarks

[feitreim]

This PR was originally a part of #53551 so theres more info about its motivation there.

• Add a CharBag prefilter on path candidates to skip irrelevant entries before invoking nucleo's matcher.
• Use binary_search on sorted matched char indices when reconstructing byte positions (perf improvement).
• Add a criterion benchmark comparing fuzzy_nucleo path matching against the existing fuzzy crate.

Performance Chart:

Benchmark	Size	Nucleo (before)	Nucleo (after)	Fuzzy	Before/Fuzzy	After/Fuzzy
1-word	100	14.14 µs	9.12 µs	9.06 µs	1.56x	1.01x
1-word	1,000	164.37 µs	114.11 µs	110.43 µs	1.49x	1.03x
1-word	10,000	1.83 ms	1.39 ms	1.41 ms	1.30x	0.99x
2-word	100	12.83 µs	3.51 µs	979 ns	13.10x	3.59x
2-word	1,000	131.65 µs	33.46 µs	6.37 µs	20.67x	5.25x
2-word	10,000	1.24 ms	338.84 µs	52.46 µs	23.64x	6.46x

Exact Current State:

query	size	nucleo	fuzzy	nucleo/fuzzy
1-word	100	8.62 µs	9.22 µs	0.93×
1-word	1000	102 µs	111 µs	0.92×
1-word	10000	1.13 ms	1.28 ms	0.88×
2-word	100	3.48 µs	0.98 µs	3.55×
2-word	1000	29.9 µs	6.39 µs	4.68×
2-word	10000	271 µs	53.4 µs	5.08×
4-word	100	0.85 µs	0.53 µs	1.60×
4-word	1000	2.99 µs	1.66 µs	1.80×
4-word	10000	20.1 µs	9.14 µs	2.20×

Self-Review Checklist:

• I've reviewed my own diff for quality, security, and reliability
• Unsafe blocks (if any) have justifying comments
• The content is consistent with the UI/UX checklist
• Tests cover the new/changed behavior
• Performance impact has been considered and is acceptable

Release Notes:

• fuzzy_nucleo: improved the performance of path matching

[feitreim]

hey @yara-blue heres Part A from the larger PR, just the path optimizations.

[feitreim]

While I was implementing the string matching into some of the existing pickers, I realized a major source of regressions was the SmartCase handling, I explain this better in #54123 but I did think it made sense to backport these changes to the path matching as well. this change is better parity with fuzzy, but at a slight perf cost. The overhead is pretty small though, maxing at about 30-60µs, even in the worst cast 10000 candidates with smart case actually on.