[BUGF-ROUTER][Fixed a small compatibility issue, and added a fallback for models without struct-output]

[IlumCI]

Description: Fix MultiAgentRouter model compatibility issues with structured outputs and JSON parsing

Issue: #1137

Dependencies: none

Tag maintainer: @kyegomez

Twitter handle: https://x.com/IlumTheProtogen

Changes made:

1. Added conditional structured outputs - only sets response_format for models that support json_schema (gpt-4.1, gpt-4o, o3-, o4-)
2. Added JSON parsing fallback for models without structured outputs - routes to first available agent when JSON parsing fails
3. Updated model names in test file to use correct LiteLLM conventions (anthropic/claude-, gemini/gemini-)

Why it didn't work last time:
The router was unconditionally setting response_format=MultipleHandOffsResponse for all models, causing BadRequestError on models like gpt-3.5-turbo that don't support json_schema structured outputs. Additionally, model names were using incorrect formats (claude-3-5-sonnet instead of anthropic/claude-3-5-sonnet-20241022) causing provider resolution failures.

Testing:

• gpt-4.1, gpt-4o, o3-mini: Success with structured outputs
• gpt-3.5-turbo: Success with JSON parsing fallback
• Anthropic/Google models: Fail only due to missing API keys (expected behavior)

Results:

===== Model Run Summary =====
gpt-4.1: ✅ Success
gpt-4o: ✅ Success
gpt-3.5-turbo: ✅ Success
gpt-4.1: ✅ Success
o3-mini: ✅ Success
anthropic/claude-3-5-sonnet-20241022: ❌ Error: litellm.AuthenticationError: AnthropicException - {"type":"error","error":{"type":"authentication_error","message":"x-api-key header is required"},"request_id":"req_011CU5WbPhYDzgh6k3TCBmym"}```


<!-- readthedocs-preview swarms start -->
----
📚 Documentation preview 📚: https://swarms--1142.org.readthedocs.build/en/1142/

<!-- readthedocs-preview swarms end -->