Bump mojo-regex to 0.9.0

[msaelices]

Summary

Bump mojo-regex from 0.7.0 to 0.9.0.

v0.9.0 - PikeVM and Lazy DFA

• PikeVM engine: Thompson NFA simulation with fixed-size SIMD state tracking and first-byte prefilter. Eliminates exponential backtracking.
• Lazy DFA: Caches PikeVM state-set transitions as DFA states. After warmup (~20 bytes), most transitions are O(1) cache hits. Patterns like flexible_phone are now 0.6x vs Rust's regex crate.
• rfind for .* last-literal: Single-pass reverse search instead of repeated forward search.
• Win rates: 96% vs Python (59/61), 53% vs Rust (30/56)

v0.8.0 - DFA and SIMD optimizations

• Nibble-based SIMD: Character class matching using native pshufb instructions
• DFA extensions: Non-capturing alternation groups, variable-length branches, deeply nested groups, capturing group flattening
• .* fast paths: Skip NFA entirely for .*LITERAL and LITERAL.* patterns
• Parser fixes: Escaped characters, dash outside brackets, DFA start state
• Fair benchmarks: Pre-compiled regex, median timing, auto-calibration across all three engines

See full changelog: https://github.com/msaelices/mojo-regex/blob/main/CHANGELOG.md