[Feature] Add knowledge base metrics methods to VectorDB

[Ervinoreo]

Is this related to a problem?
There's no easy way to query the knowledge base for statistics like document count by source, SDK version coverage, or chunk distribution. This makes it hard to monitor RAG quality or debug retrieval issues.

Describe the feature
Add diagnostic methods to the VectorDB class in src/embeddings/vectordb.py:

• get_stats_by_source() — document and chunk count per source
• get_sdk_version_coverage() — how many chunks exist per SDK version
• get_chunk_size_distribution() — min/max/avg/percentile of chunk token counts

Which module does this relate to?

• RAG / Knowledge Base

Example usage

db = VectorDB()
stats = db.get_stats_by_source()
# {'stylus-sdk': {'docs': 45, 'chunks': 312}, 'arbitrum-docs': {'docs': 23, 'chunks': 156}, ...}

coverage = db.get_sdk_version_coverage()
# {'0.10.0': 245, '0.9.2': 89, '0.6.1': 12}

Additional context

• Files: src/embeddings/vectordb.py
• ChromaDB collections support metadata queries — these methods would wrap collection queries
• Useful for the M5 final report metrics and ongoing RAG quality monitoring