Optimize Lrama parser generation performance

[ydah]

This PR optimize Lrama parser generation performance.

Metric	Before	After	Improvement
wall time	1.01713s	0.68659s	32.50%
parse parse.y	0.10654s	0.03253s	69.47%
compute_look_ahead_sets	0.35790s	0.11970s	66.55%
compute_la	0.20755s	0.01652s	92.04%

[Copilot]

🟢 Approval recommended

The changes are localized performance optimizations with preserved semantics (plus slightly safer handling in a few edge paths) and include appropriate cache invalidation.

Pull request overview

This PR focuses on reducing Lrama parser-generation time by avoiding repeated linear scans and repeated regexp compilation during grammar parsing and lookahead computation.

Changes:

• Speed up lookahead computation by iterating only existing lookback relations and caching rule lookup by id.
• Speed up state transitions by caching transitions in a symbol.number -> transition hash and using it for State#transition.
• Speed up lexing by precompiling frequently used regex patterns and reducing repeated instance-variable access in lex_c_code.

File summaries

File	Description
sig/generated/lrama/state.rbs	Adds RBS types for the new @transitions_by_symbol_number cache and accessor.
sig/generated/lrama/lexer.rbs	Adds RBS constants for the precompiled lexer regex patterns.
lib/lrama/states.rb	Optimizes compute_la to avoid scanning all rules and to reduce repeated work per state.
lib/lrama/state.rb	Adds a transition lookup cache keyed by symbol number and uses it in State#transition.
lib/lrama/lexer.rb	Precompiles regexes for symbol/percent-token scanning and refactors lex_c_code to reduce repeated work.

Review details

• Files reviewed: 3/5 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.