Improve ingestion speed for repositories with massive commits or large source code

[rycerzes]

• Details: Ingestion is very slow on large repos. Profile bottlenecks (clone, parse, chunk, embed, DB writes) and implement optimizations (parallelization/batching/caching where safe).
• Acceptance Criteria: Benchmark shows measurable end-to-end ingestion improvement on large repos, with no regression in correctness.