Modular: Support for Custom PyTorch Models

AI Impact Summary

This update introduces the ability to directly integrate custom PyTorch models into the Modular platform, expanding its capabilities beyond the standard offerings. This allows developers to leverage existing models trained on specific datasets or optimized for particular tasks, increasing flexibility and reducing the need for retraining. The change unlocks new use cases for Modular, particularly in areas requiring specialized model architectures or fine-grained control.