Add the "minmax" kernel

[AustinT]

What does this implement/fix? Explain your changes

This PR provides a memory-efficient implementation for the minmax kernel from my personal BO repo.

It uses a formulation from Ioffe (2010) that allows torch.cdist to be used instead of directly computing mins/maxes.

What testing did you do to verify the changes in this PR?

None: unfortunately I could not get the environment to build properly on my machine so I could not run the tests. Maybe you can run them if you have a pre-built environment which works? I think the environment file in the main repository is out of date...

Pull Request Checklist

• [ ] Added a note about the modification or contribution to the ./CHANGELOG.md file (if applicable)
• [ ] Added appropriate unit test functions in the ./gauche/tests/* directories (if applicable)
• [ ] Modify documentation in the corresponding Jupyter Notebook under ./notebooks/ (if applicable)
• [ ] Ran python -m py.test tests/ and make sure that all unit tests pass (for small modifications, it might be sufficient to only run the specific test file, e.g., python -m py.test tests/kernels/test_graph_kernels.py)
• [ ] Checked for style issues by running black . and isort .

[Ryan-Rhys]

Thanks very much for adding this Austin! Will run the tests on my setup!

[Ryan-Rhys]

Apologies, I just revamped the fingerprint kernel classes in line with recent changes in GPyTorch. Merged over your MinMax implementation in a separate PR. Will add a note to the file that you authored the kernel in a bit!

[AustinT]

@Ryan-Rhys can you link me to the PR where this was merged? Happy to close this PR if the functionality has been added.