[RFC] Add a Galaxy tool for TabPFN package (by Prof. Hutter's group)

[anuprulez]

The PR adds a Galaxy tool that uses the TabPFN package (DOI: 10.48550/arXiv.2207.01848) to create a general-purpose classifier.

• It takes input two files - train and test data along with their corresponding labels
• TabPFNClassifier trains on the training data and predicts classes using test data
• Additionally, it outputs a precision-recall curve.
• A conda package has been created using the TabPFN PyPi package (https://github.com/conda-forge/tabpfn-feedstock) because it is required by the Galaxy platform

Many more subsequent analyses can be added as required such as creating workflows with this tool.

A few challenges:

• One challenge is the export this classifier to standard ONNX format. The standard export of Pytorch models to ONNX does not work with the trained TabPFN model. We may require it, otherwise, we may have to export the trained model as a zipped file.

• Additionally, to move the discussion around integrating the tabpfn_client package - it requires a separate login before use, so integrating it to Galaxy would be complicated I think unless there is a way to bypass the login process.

Thank you!

ping @bgruening