Enable Ollama integration - #150

feat: Add Ollama LLM support to ADK models

This commit introduces support for using Ollama-served language models
within the Agent Development Kit (ADK). This allows developers to
easily integrate locally running LLMs (including various open-source
models) into their ADK agents, facilitating local development,
testing, and experimentation without reliance on external cloud APIs.

Rationale:

• Enable use of local and open-source LLMs with ADK.
• Provide flexibility for developers during prototyping and testing.
• Reduce dependency on cloud provider APIs for certain use cases.

Changes:

• Introduced the OllamaLlm class in src/google/adk/models/ollama_llm.py,
  inheriting from BaseLlm.
• Implements generate_content_async to communicate with the Ollama
  /api/chat endpoint using httpx.
• Supports both streaming (stream=True) and non-streaming (stream=False)
  generation.
• Includes helper functions (_convert_content_to_ollama_message,
  _build_ollama_request_payload) to convert between ADK LlmRequest/Content
  and Ollama's message format.
• Added OllamaLlm to src/google/adk/models/__init__.py and the model
  registry (src/google/adk/models/registry.py) for automatic discovery.
• Added comprehensive unit tests in tests/unittests/models/test_ollama_llm.py
  covering configuration, payload generation, streaming/non-streaming
  responses, and error handling using pytest and unittest.mock.

Dependencies:

• Introduces a dependency on the httpx library (pip install httpx)
  for asynchronous HTTP requests.

Usage:

Ensure Ollama server is running (ollama serve) and the desired model
is pulled (ollama pull <model_name>).

import asyncio
from google.adk.models import OllamaLlm, LlmRequest
from google.genai.types import Content, Part

async def run_ollama_example():
    # Model name must match a model pulled in Ollama
    MODEL_NAME = "gemma3:12b" # Adjust as needed

    try:
        # 1. Instantiate the Ollama LLM
        ollama_llm = OllamaLlm(model=MODEL_NAME, api_base="http://localhost:11434")

        # 2. Create a request
        request = LlmRequest(
            model=MODEL_NAME, # Provide model name as string
            contents=[Content(role="user", parts=[Part(text="Why is the sky blue?")])]
        )

        # 3. Generate content (non-streaming)
        print(f"Sending request to Ollama model: {MODEL_NAME}...")
        async for response in ollama_llm.generate_content_async(request, stream=False):
            if response.content and response.content.parts:
                print(f"Response: {response.content.parts[0].text}")
            else:
                print("Received empty response.")

    except Exception as e:
        print(f"Error: {e}")
        print("Ensure Ollama is running and the model is pulled.")

if __name__ == "__main__":
    # Prerequisites: ollama serve & ollama pull gemma3:12b (or your MODEL_NAME)
    asyncio.run(run_ollama_example())

See examples/multi_agent_debate.py for a more detailed example involving
multiple agents and tools.

Testing:

• The implementation is covered by unit tests located in
  tests/unittests/models/test_ollama_llm.py. These can be run using pytest.
• Manual testing was also performed using the script above and the
  multi-agent example against a running Ollama instance serving models
  like gemma3:12b.