Image Similarity with Hugging Face Datasets and Transformers

This is just applying standard deep learning patterns to an old problem. Using Hugging Face Datasets and Transformers makes it easier to grab pre-packaged similarity models, which is nice for getting started. It doesn't solve the fundamental problem of defining what 'similar' means in a complex game context. It just automates the feature extraction, which is what we do anyway, but it adds a layer of dependency management.