Local Ollama as additional embedding option #19

Local Ollama as additional embedding option

Context

Competitor uteke leads because it is 100% offline. Memayu needs to close the gap by offering more local embedding options.

Task

In addition to the Candle-embedded backend (Issue 1), add an EmbedderProvider option that points to a local Ollama instance (not Ollama Cloud) already running on the user's own machine/VPS (http://localhost:11434). This is far cheaper to implement than Issue 1 (it reuses the existing HTTP client pattern for remote providers — only the default base_url changes to localhost) and fits users who already run Ollama for other purposes.

Requirements

• Reuse the existing HttpLlmProvider/EmbedderProvider pattern in memayu-llm-client; just add a preset config provider = "ollama_local" with default base_url = http://localhost:11434
• Auto-detect dimension still applies, same as other providers

Acceptance criteria

• A user with local Ollama running (any embedding model, e.g. nomic-embed-text) can use it immediately without manually configuring base_url when on the standard port