Get hypervisor node training

[andyblarblar]

Creating the hypervisor node will involve 3 main functions:

1. Code that spawns the ros child process and kills it in the /order_66 sub
2. The data trimming algorithm
3. The actual neural network training code. We may be able to just steal this from jearock's neural_net package in OSCAR, but this will require a bit of reverse engineering to ensure what it does is valid in our enviornment.

[Bcsotty]

Initial work has been pushed to control_flow branch. Still need to test this branch prior to merge.

[andyblarblar]

Currently fails due to some weird code that attempts to crop when no crop settings are applied, and has an invalid shape on the final output layer of the default. NN. Also outputs a bunch of files in the top directory

[andyblarblar]

Need to ask jearock about the models shape being wrong. We can set the outputs to 3 and it will work, but otherwise it doesn't. I think we need to add brake to the training material to use it with 3 outputs though. Pretty sure we can just delete the stuff above line 167 in drive_train.py to solve the crash.