feat: litellm integrated for multiple llm providers

[tanmaysharma2001]

Closes #156

Summary

Replaced custom provider implementations with LiteLLM to enable unified access to multiple LLM providers (Ollama, Gemini, DeepSeek, OpenAI, Anthropic) through a single, consistent interface.

Changes made

1. Added LiteLLM Provider (models.py)

• Created LiteLLMProvider class with unified completion API
• Supports streaming, JSON format, temperature, and top_p parameters
• Automatic response format conversion for compatibility
• Configurable api_base for local providers (Ollama)

2. Updated LLM Initialization (llm_utils.py)

• Modified initialize_llm_provider() to use LiteLLM by default
• Automatic Ollama model prefixing (ollama/{model_name})
• Maintains backward compatibility with legacy providers (use_litellm=False)
• Smart provider detection based on model name

3. Enhanced Provider Support [models.py]

• Extended ModelProvider enum with:
  • DEEPSEEK
  • OPENAI
  • ANTHROPIC

4. Configuration Updates (prompt.py)

• Added model parameters for DeepSeek, OpenAI, and Anthropic models
• Added provider mappings for new models
• Added environment variable support:
  • DEEPSEEK_API_KEY
  • OPENAI_API_KEY
  • ANTHROPIC_API_KEY
  • OLLAMA_API_BASE (default: http://localhost:11434)