fix: align default LLM model with registry

[wuyua9]

Summary

• add a shared DEFAULT_MODEL_NAME for the checked-in API model registry
• replace stale gpt-4.1 defaults in backend request schemas, backtesting service, CLI run path, and call_llm fallback with the registered default
• update the frontend default selector and add a regression test so runtime defaults cannot drift away from api_models.json

Notes

This intentionally does not change the OPENAI/OpenAI provider fallback because that is already covered by #511.

Tests

• python3 -m pytest tests/test_llm_default_model.py -q
• git diff --check origin/main...HEAD