Move new model builders from Prototype to main TorchVision

[datumbox]

🚀 The feature

Once all model builders are updated to support multiple pretrained weights, we should replace the old builders with the new ones. This involves the following steps:

1. Review the naming conventions for final time based on [FEEDBACK] Multi-weight support prototype API #5088. Also decide on the naming convention of _presets.py (ImageNetEval or ClassificationEval?)
2. Port to torchvision.models the inplace code changes of torchvision.prototype.models. Porting depends on having agreement on the deprecation of positional arguments across the entire TorchVision. If such agreement doesn't exist at the time of porting, we will need to remove the mandatory keyword arguments from the API prior landing it to the main area.
3. Delete torchvision.prototype.models and their prototype tests.
4. Update the existing model tests (pretrained => weights) and ensure all work without modifying the expected files.
5. Update the reference scripts to work with the new API.
6. Update the documentation of the new model builder methods.
7. Update the tutorials and gallery with how to use the pretrained models and do preprocessing.

cc @datumbox @pmeier @bjuncek