perf(treesitter): compile AST rules into lookup sets and cache pointers

[HarshK97]

Problem

1. When building an AST, buildASTWithRules was running slices.Contains across 5 different slices (Ignored, Keywords, LabelIgnored, Unordered, Flattened) on every single node, so on larger files with thousands of nodes, doing linear scans over and over added unnecessary overhead.
2. AreTypesEquivalent was looping through nested 2D slices on every node comparison during matching.
3. GetRules stored and returned Rules by value, so it copied the whole struct on every language lookup.

What Changed

• Precomputed map[string]struct{} lookup sets for Flattened, Ignored, LabelIgnored, Keywords, Blocks, and Unordered at startup in compileSets(), turning linear slice checks into O(1) map lookups while keeping slice fallback for uncompiled rules.
• Indexed EquivalentTypes into an equivGroups map (map[string][]int), so type equivalence lookups just check matching group IDs directly instead of scanning every group.
• Added nil-safe helper methods on *Rules (IsIgnored, IsKeyword, IsLabelIgnored, IsUnordered, IsFlattened, IsBlock, Alias) to keep all rule lookups clean and encapsulated in buildASTWithRules.
• Changed rulesCache to store and return *Rules pointers directly, stopping all the struct copying on GetRules calls.
• Added an isomorphic fast path in engine.Match so identical subtrees pair up immediately when root node hashes match.
• Added unit tests in internal/treesitter/rules_test.go covering compiled sets, uncompiled fallback, and nil receiver safety across all helper methods.