Check in PyTorchModel

[kathryn-baker]

Implementation of a PyTorchModel class. Currently allows loading model via python objects or from yaml, although loading from yaml requires assigning transformation objects after object creation

[roussel-ryan]

@ChristopherMayes are we ok with the file overhead of testing on the California dataset?

[roussel-ryan]

its certainly aiding in creating a robust testing suite

[kathryn-baker]

A summary of what I've found out relating to the conda-build stage:

• the reason the build was being cancelled originally is because it couldn't solve the environment so I've removed the tensorflow and botorch requirements for the conda tests
• I've added an importorskip statement to the keras/torch tests so they won't run in the conda build stage but they should run in the pip stage, which will prevent the conda build from running if they fail so it's not a big loss in the test suite
• I've done a similar thing for any FileNotFoundErrors, any tests that require files in the conda build stage are skipped but should still be run in the pip stage
• I've kept the setup.py install call in the build.sh despite the warnings that this method is deprecated because using pip threw an error when I ran it locally and most other example recipes I've seen use the setup.py call