Revert the embedder to bge-small: the bge-base A/B showed no gain

Swap the embedder: bge-small (384d) -> bge-base (768d)

Codebase quality pass: kill silent-truncation hacks, fix latent bugs, cover them (#46)

Switch embeddings to a local CPU model (bge-small): free tier APIs can't index the corpus

Make Gemini the default generation provider (free tier)

Add M0 smoke script: Neon write/read, one LLM call, one Gemini embedding