feat: add code context metadata foundation

[raoabinav]

Evaluator Summary

• Abi added an AST-aware code-context metadata layer for Python chunks, capturing module paths, qualified symbols, definitions, imports, calls, and references so LEANN can reason over code structure instead of only raw text.
• The design keeps extraction pure-Python and metadata-only in this PR: low-risk, serializable, and suitable as a stable contract for later related-code expansion and cross-file symbol traversal.
• The implementation integrates at chunking time, so downstream search, MCP, and agent workflows can consume richer code metadata without changing the vector backend API.
• Quality bar: focused tests cover code-context extraction, AST chunk integration, RAG app no-recompute behavior, and chat propagation; lint/format checks were run on the changed modules.

[raoabinav]

Replaced by #354 from the neutral feature branch.