Add first-class Ollama preset

[yuanhao]

Closes #37.

Summary

• Add first-class ModelConfig::ollama(base_url, model_id) support.
• Honor OpenAiCompat::requires_assistant_after_tool_result when serializing OpenAI-compatible tool result runs.
• Keep ModelConfig::local(...) neutral for LM Studio, llama.cpp, vLLM, and other compatible servers.
• Document Ollama alongside the other first-class model presets.

Compatibility

• Existing serialized OpenAiCompat configs remain compatible because the new flag defaults to false during deserialization.
• Existing local OpenAI-compatible behavior is unchanged unless callers opt into the Ollama preset or flag.

Validation

• cargo fmt --check
• cargo test provider::model
• cargo test