feat: multi-language support via tree-sitter (Python, Go)

[bntvllnt]

Summary

Add tree-sitter-based parsing for Python and Go alongside the existing TypeScript Compiler API parser. TypeScript-only support locks out 80%+ of the market.

Motivation

• codebase-memory-mcp supports 64 languages via tree-sitter — single Go binary
• CKB uses SCIP indexes for multi-language support
• CodePathfinder is Python-only but proves demand for non-TS analysis
• Every competitive analysis perspective flagged single-language as the existential risk
• Python is the most common language in AI agent codebases

Approach

Phase 1: Parser Abstraction

• Extract parser interface from current TS Compiler API implementation
• Define ParsedFile contract that both parsers satisfy
• Keep TS Compiler API for TypeScript (higher quality: type resolution, call sites)

Phase 2: tree-sitter Integration

• Add tree-sitter + language grammars as dependencies
• Implement parser for Python (functions, classes, imports, exports)
• Implement parser for Go (functions, structs, imports, packages)
• Map tree-sitter AST nodes to existing ParsedFile interface

Phase 3: Graph Unification

• Multi-language files in same graph
• Cross-language edges (e.g., Python calling TS via API)
• Metrics compute identically regardless of source language

Tradeoffs

Aspect	TS Compiler API	tree-sitter
Type resolution	Full	None
Call site confidence	type-resolved	text-inferred only
Language support	TypeScript only	64+ languages
Parse speed	Slower	Faster
Dependency	typescript npm	tree-sitter + grammars

Acceptance Criteria

• Python files parsed: functions, classes, imports, exports
• Go files parsed: functions, structs, imports, packages
• Mixed-language repos produce unified graph
• All existing metrics compute on non-TS files
• TS Compiler API still used for .ts/.tsx files (no quality regression)
• Tests for Python and Go parsing with real fixture files

Priority

Short-term — Critical for market viability. Start with Python.