Embeddings support (OpenAI & OpenAI‑compat)

[evanmschultz]

I’d like to add embeddings support to Fantasy with a non‑breaking API surface.

Scope (v1):

• OpenAI embeddings, plus OpenAI‑compatible providers (OpenRouter, Azure, Together, etc.) via the existing
  OpenAI‑compat layer.
• No base64 output in v1 (float vectors only).
• Core API is additive: EmbeddingProvider is a separate interface; existing Provider remains unchanged.

Proposed API:

• EmbeddingProvider.EmbeddingModel(ctx, modelID)
• EmbeddingModel.Embed(ctx, EmbeddingCall)
• Core types: EmbeddingCall, EmbeddingResponse, Embedding (vectors are []float32).

Notes:

• Inputs validation: exactly one of Input or Inputs, and non‑empty.
• Convert OpenAI’s []float64 embeddings to []float32 at the provider boundary.
• Tests included for validation and OpenAI embeddings path.