fix: BoundedBacktracker span-based CanHandle + ReplaceAllStringFunc O(n)

[kolkov]

Summary

• BoundedBacktracker span-based CanHandle — SearchAtWithState(haystack, at, state) now checks CanHandle(len(haystack) - at) instead of CanHandle(len(haystack)). The visited table is sized for the search span only, with positions stored relative to SpanStart. Full haystack is preserved for \b assertions. This matches Rust regex's Input span model (backtrack.rs line 1848).
• ReplaceAllStringFunc O(n) performance — Replaced result += string O(n²) concatenation with strings.Builder. On 150K replacements over 6MB string: 2m19s → 1.3s.
• Tests — Added TestBoundedBacktracker_SearchAtWithState_SpanBased (large haystack span test) and TestBoundedBacktracker_SearchAtWithState_WordBoundary (\b context preservation).

Fixes #127. Reported by @kostya via #124.

Test plan

• go test ./... — all 9 packages pass
• gofmt -l . — no formatting issues
• Verified LogParser: 13 patterns × 52000 lines (7MB) — all match counts identical to stdlib
• Verified Template: ReplaceAllStringFunc 150K replacements — results identical, 1.3s vs 2m19s
• Span-based test: haystack > maxInputSize, search from late position finds matches
• Word boundary test: \bfoo\b with search starting mid-haystack preserves backward context
• CI: tests + benchmarks + lint

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[github-actions]

Benchmark Comparison

Comparing main → PR #128

Summary: geomean 121.9n 121.8n -0.06%

⚠️ Potential regressions detected:

geomean                               ³                +0.00%               ³
geomean                               ³                +0.00%               ³
geomean                         ³                +0.00%               ³
geomean                         ³                +0.00%               ³
AhoCorasickLargeInput/coregex_IsMatch_64KB-4            109.4µ ± ∞ ¹    116.2µ ± ∞ ¹    +6.25% (p=0.008 n=5)
MatchAnchoredLiteral/no_match_prefix-4                  2.500n ± ∞ ¹    2.508n ± ∞ ¹    +0.32% (p=0.040 n=5)
ASCIIOptimization_Issue79/short_WithASCII-4             276.1n ± ∞ ¹    279.2n ± ∞ ¹    +1.12% (p=0.008 n=5)
ASCIIOptimization_Issue79/medium_WithoutASCII-4         1.646µ ± ∞ ¹    1.692µ ± ∞ ¹    +2.79% (p=0.008 n=5)
ASCIIOptimization_Issue79/long_WithASCII-4              1.101µ ± ∞ ¹    1.202µ ± ∞ ¹    +9.17% (p=0.032 n=5)
ASCIIOptimization_Issue79/long_WithoutASCII-4           2.638µ ± ∞ ¹    2.791µ ± ∞ ¹    +5.80% (p=0.008 n=5)

Full results available in workflow artifacts. CI runners have ~10-20% variance.
For accurate benchmarks, run locally: ./scripts/bench.sh --compare