Model usage statistics for cleanup suggestions

### Feature Idea

**Model Usage Statistics & Cleanup Assistant**

I suggest adding a built-in model usage tracking system to ComfyUI. Currently, after downloading a model to test a workflow, there's no way to know if it's still being used. Over time, the models folder accumulates hundreds of GB of unused models.

**Proposed Solution:**

1. Add a lightweight SQLite database (`model_stats.db`) in the ComfyUI root directory.
2. Track each model file when it's loaded: record model_path, model_type, file_size, first_used timestamp, last_used timestamp, and use_count.
3. Insert a one-line hook in every loader function (CheckpointLoaderSimple, LoraLoader, VAELoader, UNETLoader, CLIPLoader, ControlNetLoader, etc.) to record usage when a model is loaded.
4. Provide a simple UI panel (e.g., a custom node or a tab in the main UI) that displays:
- Model file name
- Type (checkpoint/lora/vae/controlnet/etc.)
- File size
- Last used date
- Usage count
- A "safe to delete" suggestion
5. Allow sorting by last_used or usage_count, so users can easily identify models that haven't been used for months and can be safely deleted.

This is a framework-level feature, not a custom node, because it requires hooks in all built-in loader nodes. It would greatly improve the user experience for anyone who frequently tests workflows.

### Existing Solutions

I searched for existing custom nodes that track model usage, but found none. The only workaround is manually maintaining a readme.md file, which is error-prone and unsustainable. Some users use file system last-modified timestamps as a rough proxy, but this only tracks download time, not actual usage.

### Other

Currently my models folder is ~600GB and I cannot tell which models are safe to delete. This feature would solve a real pain point for many users.

Start by mapping the named loader entry points—CheckpointLoaderSimple, LoraLoader, VAELoader, UNETLoader, CLIPLoader, and ControlNetLoader—and the main UI or custom-node entry point. Define how model_stats.db records each load and how the panel presents sorting and deletion suggestions; done means built-in loaders report usage and the UI displays the requested statistics.

Written by the indexing model from the issue text.