Restructure Abstract UDF classes to allow for the future addition of EVA transforms

[LordDarkula]

Create AbstractUDF base class and update AbstractClassifierUDF base class to inherit from it
Create AbstractTransformationUDF base class
Create PytorchAbstractTransformationUDF to combine torchvision transforms and AbstractTransformationUDF
Rename PytorchAbstractUDF to PytorchClassifierAbstractUDF

[jarulraj]

do we need input format and label properties for every UDF -- that is leading to a slight drop in coverage in #303 ?

[jarulraj]

@LordDarkula @gaurav274 Is it possible to use some kind of reflection to use torchvision transforms instead of writing a separate UDF for each transform? -- https://stackoverflow.com/questions/37628/what-is-reflection-and-why-is-it-useful

[LordDarkula]

@LordDarkula @gaurav274 Is it possible to use some kind of reflection to use torchvision transforms instead of writing a separate UDF for each transform? -- https://stackoverflow.com/questions/37628/what-is-reflection-and-why-is-it-useful

I assume you mean allow the user to just use a torchvision transform directly in EVA instead of having to explicitly subclass PytorchAbstractUDF. Assuming we want EVA to be type-safe, we could use a Protocol (https://peps.python.org/pep-0544/). However, this feature was introduced in Python 3.8 and EVA supports earlier versions of Python.

[jarulraj]

This looks more like the approach I suggested:

1. https://docs.bentoml.org/en/v1.0.0-a7/frameworks/easyocr.html
2. https://docs.bentoml.org/en/0.13-lts/frameworks.html#bentoml.frameworks.easyocr.EasyOCRArtifact