Support dataset-configurable indexing and embedding

[mikewaters]

Blocks: Support multiple content types #56

Support per-dataset indexing strategy

• In Obsidian, different document types (per ontology) may require unique chunking strategies; a landscape document for example, or a logbook, require something more advanced (split by h3 for example).

• Different types of markdown content may require different splitting

• Different content types definitely require their own indexing

• Define IndexingStrategy

• Support more indexing methods for broader content storage and for better markdown search

Support per-dataset embedding model.

Requires the vector store can do this, and upon retrieval the app uses the right model.

[mikewaters]

Research: Semantic Chunking https://kudra.ai/metadata-enriched-retrieval-the-next-evolution-of-rag/