fix: use wildcard for llm.model_name in MCP trace tests

[Chibionos]

Summary

Replaces hardcoded LLM model version with wildcard (*) in trace test expectations to prevent future test failures when models are updated.

Problem

The simple-local-mcp test expectations were recently updated to use gpt-4.1-mini-2025-04-14, but this approach is brittle - it will break again the next time:

• A new model version is released
• LLM Gateway defaults change
• Test environments use different model configurations

Solution

Use wildcard matching for llm.model_name instead of exact version:

- "llm.model_name": "gpt-4.1-mini-2025-04-14"
+ "llm.model_name": "*"

Also removed exact content matching for the final response, as wording can vary slightly between models.

Benefits

✅ Future-proof: Won't break on model updates
✅ Environment-agnostic: Works regardless of which model is configured
✅ Lower maintenance: No need to update test expectations when models change
✅ Still validates: Provider (azure), system (openai), and span structure are still checked

Implementation

The trace assertion logic (trace_assert.py) already supports wildcards:

def matches_value(expected_value: Any, actual_value: Any) -> bool:
    if expected_value == "*":
        return True  # Accept any value

Testing

This change only affects test expectations, not runtime behavior. The wildcard will accept any model name while still validating:

• Correct span structure and hierarchy
• Required span attributes
• Message roles and content
• Tool invocations

🤖 Generated with Claude Code