Add Ollama local LLM provider support

[SL-Mar]

Summary

This PR adds support for Ollama, enabling users to run large language models locally without API costs. The implementation includes a new OllamaProvider class and integrates it into the existing LLM factory pattern.

Key Changes

• New OllamaProvider class (quantcoder/llm/providers.py):

  • Implements local LLM inference using Ollama's OpenAI-compatible API
  • Supports multiple models (llama3.2, codellama, qwen2.5-coder, etc.)
  • Configurable base URL for custom Ollama deployments
  • Async chat completion support consistent with other providers

• Updated LLMFactory (quantcoder/llm/providers.py):

  • Added "ollama" to PROVIDERS registry
  • Made api_key parameter optional (not required for Ollama)
  • Added base_url parameter to support local provider configuration
  • Updated factory method to handle Ollama's different initialization requirements

• Configuration updates (quantcoder/config.py):

  • Added ollama_base_url and ollama_model fields to ModelConfig
  • Updated provider comment to include "ollama" option
  • Exported Ollama settings in to_dict() method for configuration serialization

• Bug fix: Fixed typo in LLMFactory.PROVIDERS ("Mistral Provider" → "MistralProvider")

Implementation Details

• Ollama provider uses OpenAI's AsyncOpenAI client with a custom base URL, leveraging Ollama's OpenAI-compatible API endpoint
• Default configuration points to http://localhost:11434/v1 with llama3.2 model
• API key is set to "ollama" (required by OpenAI client but unused by Ollama)
• Maintains consistency with existing provider interface for seamless integration