Implement Longest() method for stdlib compatibility

[kolkov]

Summary

The Longest() method is currently a no-op stub. For full stdlib regexp compatibility, it should switch matching semantics from leftmost-first to leftmost-longest.

Current behavior

re := coregex.MustCompile(`(#|#!)`)
re.Longest()
result := re.ReplaceAllString("#!a", "")
// Returns: "!a" (leftmost-first, matches "#")
// Expected: "a" (leftmost-longest, matches "#!")

Expected behavior (stdlib compatible)

// Default: leftmost-first (Perl semantics)
re := regexp.MustCompile(`(#|#!)`)
re.ReplaceAllString("#!a", "") // "!a"

// After Longest(): leftmost-longest (POSIX semantics)
re.Longest()
re.ReplaceAllString("#!a", "") // "a"

Research findings

Engine	Default	Longest() support
Go stdlib	leftmost-first	✅ Yes
Rust regex	leftmost-first	❌ No
RE2	leftmost-first	❌ No
coregex	leftmost-first	❌ No (stub)

Note: Neither Rust regex nor RE2 implement leftmost-longest. However, for true stdlib drop-in compatibility, we should support it.

Implementation plan

1. Add longest bool flag to Regex struct
2. Modify Longest() to set the flag
3. Update PikeVM search to continue looking for longer matches when flag is set
4. Propagate flag through meta engine coordination
5. Benchmark to ensure no performance regression in default mode

Performance considerations

• Default mode: Expected ~0% overhead (single bool check)
• Longest mode: Expected 10-50% overhead (must check all alternations)

Acceptance criteria

• Longest() actually switches to leftmost-longest semantics
• Default mode performance unchanged (verified by benchmarks)
• All stdlib Longest() behaviors matched
• Documentation updated

Related

• Discovered via GoAWK integration: DRAFT: test the coregex library benhoyt/goawk#264
• AWK uses POSIX (leftmost-longest) semantics by default