fix: sync RLM model/provider into child calls after interactive model switch

[GitHubFoxy]

Bug

When starting a session in ypi with ChatGPT model, hitting a limit, and switching to another model, rlm_query still uses the ChatGPT model. Expected: rlm_query should use the newly selected model, same as the parent.

Fix

This change ensures that when a user switches models interactively, the RLM_PROVIDER and RLM_MODEL environment variables are properly synchronized into child calls.

Key improvements:

• Extracted syncRlmEnv and updateUi helper functions for better code organization
• Added model_select event handler to update environment when user changes models
• Preserved existing model state on session_start with preserveExistingModel option
• Ensures bash tool executions inherit correct model information

Testing

Verified by:

1. Starting ypi session with one model
2. Switching to a different model interactively
3. Confirming rlm_query uses the new model for sub-calls