File name handling for model loading is too rigid — requires unnecessary manual renaming

### Feature Idea

Hi,

I’m encountering issues with the current model loading mechanism, especially for IPAdapter and LoRA models, but this applies to others as well (e.g., CLIP backbones).

The problem is that the loader expects very specific filenames (like faceid.plus.sd15.lora.safetensors), but most official sources (like Hugging Face or Civitai) provide files with very different names. For example:
ip-adapter-faceid_sd15_lora.safetensors → requires renaming

CLIP variants:
CLIP-ViT-H-14-laion2B-s32B-b79K.safetensors
CLIP-ViT-bigG-14-laion2B-39B-b160k.safetensors
→ both are named model.safetensor upon download

This creates unnecessary friction: users are forced to manually rename files to match hardcoded patterns in the codebase, or else the models silently fail to load.

I suggest improving this in one of the following ways:
Make the file-matching logic more flexible (e.g. allow known variants or fallback matches).
Provide clearly named downloadable files that already match the expected naming scheme.
Alternatively, allow manual file selection in the UI instead of hiding it behind strict presets.
This would save a lot of time and confusion, especially for new users.

Thanks.

### Existing Solutions

_No response_

### Other

_No response_