perf: PikeVM sparse-dispatch for dot patterns — 2.8-4.8x speedup (#132)

[kolkov]

Summary

• NFA compiler generated ~9 split states chaining UTF-8 byte-range alternation branches for each . (AnyCharNotNL). PikeVM had to DFS-traverse the entire split chain at every byte position — O(branches) work per byte.
• New compileUTF8AnySparse() compiles . as a single sparse state mapping each leading byte range directly to its continuation chain — O(1) dispatch instead of O(branches) split-chain traversal. Same approach as Rust regex's State::Sparse.
• Sparse NFA (runeNFA) compiled only for patterns containing .. PikeVM and BoundedBacktracker use it; DFA continues with byte-level NFA (unaffected).
• PikeVM speedup: 2.8-4.8x on dot-heavy patterns (.*?, .+, .*).

Reported by @kostya via LangArena benchmarks (Issue #124). Closes #132.

Test plan

• go test ./... — all 11 packages pass
• gofmt -l . — clean
• golangci-lint run — 0 issues
• BenchmarkFind — no regressions
• Full meta benchmark suite — PASS, no regressions
• CI: tests + benchmark comparison
• regex-bench validation (post-merge)

[github-actions]

Benchmark Comparison

Comparing main → PR #134

Summary: geomean 121.6n 119.0n -2.08%

⚠️ Potential regressions detected:

LazyDFASimpleLiteral-4     59.95n ± ∞ ¹   62.21n ± ∞ ¹  +3.77% (p=0.008 n=5)
LazyDFAAlternation-4       66.12n ± ∞ ¹   66.79n ± ∞ ¹  +1.01% (p=0.008 n=5)
LazyDFARepetition-4        85.48n ± ∞ ¹   87.51n ± ∞ ¹  +2.37% (p=0.008 n=5)
geomean                               ³                +0.00%               ³
geomean                               ³                +0.00%               ³
geomean                         ³                +0.00%               ³
geomean                         ³                +0.00%               ³
AhoCorasickManyPatterns/coregex_10_patterns-4           59.16n ± ∞ ¹    67.34n ± ∞ ¹   +13.83% (p=0.008 n=5)
AhoCorasickManyPatterns/coregex_25_patterns-4           72.55n ± ∞ ¹    81.59n ± ∞ ¹   +12.46% (p=0.008 n=5)
AhoCorasickManyPatterns/coregex_50_patterns-4           88.08n ± ∞ ¹    91.52n ± ∞ ¹    +3.91% (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

[codecov]

Codecov Report

❌ Patch coverage is 44.57831% with 46 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
nfa/compile.go	0.00%	38 Missing and 2 partials ⚠️
meta/compile.go	86.04%	5 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!