Inference execution for Candil.

Handles chat completions and embeddings for both local engines and remote providers. Normalises the request/response format across the supported provider APIs (OpenAI, Anthropic, Ollama, OpenAI-compatible).

Message format

All messages are plain maps with :role and :content string keys:

%{role: "system", content: "You are a helpful assistant."}
%{role: "user", content: "Hello!"}
%{role: "assistant", content: "Hi there!"}

Response format

All chat functions return a response() map:

%{
  content: "Hello, how can I help?",
  role: "assistant",
  model: "llama-3-8b",
  finish_reason: "stop",
  usage: %{prompt_tokens: 12, completion_tokens: 8, total_tokens: 20}
}