feat: Ollama embedding support

[TerminalGravity]

Closes #6

Adds Ollama as a third embedding provider alongside local (Xenova) and OpenAI.

Changes

• OllamaEmbeddingProvider class using Ollama's /api/embed endpoint
• Native batch embedding support (single API call for multiple texts)
• Configurable: base URL, model name, dimensions
• Defaults: nomic-embed-text (768 dims) on localhost:11434
• Env vars: EMBEDDING_PROVIDER=ollama, OLLAMA_BASE_URL, OLLAMA_EMBED_MODEL
• Updated config types, timeline-db, and onboard-project tool
• 3 new unit tests for the Ollama provider factory

Usage

# Just set the env var (Ollama must be running)
EMBEDDING_PROVIDER=ollama preflight

# Or in .preflight.json
{ "embeddings": { "provider": "ollama" } }

# Custom model
OLLAMA_EMBED_MODEL=mxbai-embed-large preflight

Zero external API keys needed — perfect for local-first users.

[TerminalGravity]

This is a great addition — clean implementation, native batch support, and sensible defaults (nomic-embed-text, localhost:11434). A couple thoughts:

1. Dimension mismatch handling — if someone switches between providers mid-project (e.g. local → ollama), the stored embeddings will have different dimensions. Might be worth a warning or migration note in the docs.

2. Timeout/retry — the Ollama fetch calls don't have a timeout. If the local server is down, these will hang. Consider an AbortController with a reasonable timeout (10-15s).

3. Relation to #81 — this overlaps significantly with PR #81 (feat: add Ollama as local embedding provider). Are these meant to supersede it, or should one be closed?

Otherwise looking good. Would approve once the overlap with #81 is clarified.

[TerminalGravity]

Superseded by #148.