fix: Scope boundaries for JS/TS entity extraction

[c22]

The entity extractor recursed into JS/TS function expression bodies (arrow functions, function expressions, generator functions) and extracted local variables (const, let, var) as top-level entities. These spurious entities are unstable and small structural changes to surrounding code could cause different locals to be extracted, producing different entity sets from logically equivalent code.

Downstream consumers such as weave would interpret the instability as intentional additions or deletions, which could cause code to be silently dropped or mangled during merge conflict resolution.

Scope boundaries are now selectively transparent: local variable declarations inside function expression bodies are suppressed, but inner class and function declarations are still extracted as entities.

This keeps entity extraction stable across versions while preserving granularity for real semantic units.

Implementation uses two mechanisms configured through LanguageConfig:

1. scope_boundary_types When the general recursion in visit_node encounters one of these node types, it propagates the boundary as the suppression_context rather than skipping the subtree entirely.

2. suppressed_nested_entities Rules keyed on suppression_context filter out lexical_declaration and variable_declaration inside scope boundary types and named function/method bodies.

Both are configured through LanguageConfig fields and currently only apply to JavaScript/TypeScript.