Chunk long notes before embedding

[BeastOfShadow]

Problem

sync_notes() and the ingest pipeline add whole documents to ChromaDB
(collection.add(documents=[content], ...)). Long notes become a single oversized
vector, diluting semantic precision and hurting retrieval.

Proposal

Split note content into overlapping chunks before embedding, so retrieval returns
focused passages instead of whole files.

Tasks

• Add a chunker (e.g. ~500–800 tokens, ~10–15% overlap; split on headings/paragraphs first)
• Apply it in src/database/vector_db.py (sync) and src/pipeline/orchestrator.py (ingest)
• Store source + chunk index in metadata so sources still map back to the origin note
• Update sync_notes() delete/update logic to handle multiple chunk-ids per file
• Make chunk size / overlap configurable via .env

Acceptance

A long note is indexed as multiple chunks; a query returns the relevant passage, and sources still resolve to the correct file.