Add support for loading compiled `.mlmodelc` models

[gavrilikhin-d]

Currently, Swift version of UForm always assumes that the model is uncompiled, which isn't a case, if you bundle it together with your app

[hbcondo]

+1 for this. unum-cloud's own models successfully compile to .mlmodelc:

xcrun coremlc compile uform3-image-text-english-small/text_encoder_neural.mlpackage ./CompiledModel

But perhaps another way of requesting this feature is to add an "offline mode" for uform that revokes network connectivity. The .mlmodelc file can be attached to the swift project so embedding can be done on-device. But to be truly offline, the TextEncoder would also need to be updated as it currently just accepts a modelPath parameter.

Overall, this would make for a privacy-first solution and could possibly be implemented similar to Hugging Face's swift-transformers library offline mode introduced this year:
huggingface/swift-transformers#177

[gavrilikhin-d]

#108