Add the ability to specify other types of resources

[FrankD412]

This issue address the need to be able to specify resources that are not the standard set of processors and nodes through a scheduler. Those resources include (but are not limited to) GPUs, ASICs, and other forms of scheduler assignable hardware. The items that this issue should address at the very core are as follows:

• Addition of specification notation for requesting resources.
• New logic to existing abstract and concrete scheduler adapters for formatting the request.

A potential format for the YAML study step could be as follows:

- name: legacy-token-prepend
      description:
          The legacy format for simply appending all of the step's resources to
          a single command. Procs is the only resource required for this format.
      run:
          cmd: |
            $(LAUNCHER) echo "Test print #1"
          depends: []
      resources:
          nodes: 1
          procs: 1
          gpus: 2

The idea would be that the resources dictionary is kind of a catch-all for resources. Each adapter would be responsible for throwing an exception if specified resources are not capable of being requested with a particular scheduler.

[FrankD412]

The more I think about this format, the more I like it. This change would necessitate a v2.0 of the YAML study description, Thinking that a spec revamp is going to be a project.

[gonsie]

Here is a use case to think about (though it may fit better with #77):

Burst buffer on Sierra will be scheduled via command line arguments to bsub. Via this argument, users can specify a stage-in script, 2 stage-out scripts, and the amount of node-local storage space to request. I can point to some internal confluence documentation (once I'm done writing it).

[FrankD412]

@gonsie -- I may need to sit down with you and look at how this would look. And definitely pass along the documentation when it's ready!

[FrankD412]

This feature goes hand in hand with the specification redesign. Pushing this to a later release.