feat: Add IBM WatsonX AI client integration

[injusticescorpio]

Summary

This PR adds support for IBM WatsonX AI as a new backend for RLM, enabling users to run language models through IBM's WatsonX AI platform. The integration uses LangChain's ChatWatsonx client, providing tool calling support, streaming capabilities, and structured output handling.

Implemented:

• Complete WatsonX client implementation with sync and async support
• Message format conversion (string, dict, and multi-turn conversations)
• Usage tracking (tokens and call counts)
• Environment variable configuration support
• Comprehensive test suite with 100% coverage of core functionality
• Working example demonstrating usage
• Integration into RLM's backend routing system

Note: WatsonX API doesn't provide cost information in responses hence it returns None.

Test Plan

• Unit tests added covering:
  • Message format conversion (string, dict, list)
  • Sync and async completion methods
  • Error handling for invalid inputs
• All tests pass with mocked WatsonX client
• Example code provided for manual testing with real credentials
• Integration follows existing client patterns (OpenAI, Anthropic, Gemini)

Breaking Changes

None. This is a purely additive change.

[injusticescorpio]

@alexzhang13 could you please review and merge this PR?

[injusticescorpio]

@alexzhang13 could you please review and merge this PR?