fix: llama.cpp provider should use list() result instead of cached models list #5859
Description
Description
The llama.cpp provider currently uses a cached models list instead of calling the list() method to get the current state of available models. This causes issues when users manually delete model files or directories - the provider continues to show deleted models in the UI, creating confusion and potential errors when users try to load non-existent models.
The cached list becomes stale and doesn't reflect the actual filesystem state, leading to inconsistencies between what's displayed and what's actually available.
Expected Behavior
The llama.cpp provider should always use the result from the list() method to display available models. This ensures:
- Models list always reflects the current filesystem state
- Manually deleted models are immediately removed from the UI
- No stale cached data causing incorrect model availability
- Consistent and accurate model listing across all provider operations