Add Ollama executor support for local model inference

[Darktex]

This commit adds full Ollama integration to LLMVM, enabling users to run
local language models with complete tool calling support using the
block pattern.

Key Features:

• New OllamaExecutor that inherits from OpenAIExecutor
• Leverages Ollama's OpenAI-compatible API endpoint
• Full conversation mode support
• Tool calling via blocks for local code execution
• Supports popular models: llama3.1, qwen2.5, mistral, etc.

Implementation Details:

• llmvm/common/ollama_executor.py: Main executor implementation

  • Inherits from OpenAIExecutor for OpenAI compatibility
  • Configurable endpoint (default: http://localhost:11434/v1)
  • No API key validation (Ollama runs locally)
  • Supports streaming responses and stop tokens

• llmvm/common/helpers.py: Executor registration

  • Added 'ollama' executor to get_executor() method
  • Environment variable support: OLLAMA_API_BASE
  • Config variable: ollama_api_base, default_ollama_model

Testing & Documentation:

• scripts/test_ollama.py: Comprehensive test suite

  • Tests conversation mode
  • Tests tool calling with blocks
  • Validates model availability
  • Supports multiple models and endpoints

• docs/OLLAMA.md: Complete setup and usage guide

  • Installation instructions
  • Configuration options
  • Model recommendations
  • Troubleshooting guide
  • Performance tips

• README.md: Updated to mention Ollama support

Usage:
LLMVM_EXECUTOR='ollama' LLMVM_MODEL='llama3.1' python -m llmvm.client

Testing:
python scripts/test_ollama.py

Requirements:

• Ollama installed and running (https://ollama.ai)
• A model pulled (e.g., ollama pull llama3.1)

Closes requirement: Enable Ollama compatibility with conversation mode
and full tool calling support using blocks.