Title: Add backend unit tests for LangGraph agent

Description:

The backend currently has no automated tests. We need comprehensive unit tests to verify the LangGraph agent behaves correctly.

Requirements:

1. Test infrastructure setup

   • Add pytest with async support for testing async agent operations
   • Add respx for mocking HTTP requests to OpenAI-compatible APIs
   • Add a moon backend:test task to run tests

2. Test scenarios to cover

   • Basic conversation: Agent receives a user message and returns an AI response
   • State maintenance: Conversation history is preserved across multiple invocations within the same thread
   • Tool invocation: Agent recognizes when to call the get_weather tool, executes it, and incorporates the result into its response
   • Tool output verification: The tool returns the expected format and content

3. Test parametrization

   • Tests should run against multiple OpenAI-compatible base URLs (OpenAI default, OpenRouter, Ollama) to ensure provider flexibility
   • Tests should run with different model names to ensure model configuration works correctly

4. Determinism requirements

   • The get_weather tool has random sleep for realism in production, but tests must be fast and deterministic
   • Mock the tool to return immediately with predictable output

Acceptance criteria:

• moon backend:test passes
• Tests cover all four scenarios above
• Tests are parametrized across providers and models
• No flaky tests due to randomness or timing