Override constrain_resources() in individual steps

[xylar]

This merge drops calls to Testcase.run() and instead override step.constrain_resources() in individual steps.

We have deprecated Testcase.run() because its use is not compatible with the task parallelism approach we (@altheaden and I) are developing. Instead, the individual steps in a test case that need to set the number of tasks or cpus per task should override constrain_resources(), set self.ntasks, self.min_tasks, etc., and then call super().constrian_resources().

In the near future (again related to task parallelism), it will also be useful to override step.runtime_setup() for tasks like creating graph partitions before running an MPAS component. In anticipation of this change and to be consistent with the deprecation warning about Testcase.run(), this merge adds a call to runtime_setup() when a step gets run. Currently, no steps override this function.

Checklist

• Developer's Guide has been updated
• API documentation in the Developer's Guide (api.rst) has any new or modified class, method and/or functions listed
• Document (in a comment titled Testing in this PR) any testing that was used to verify the changes

[xylar]

Testing

I ran the pr test suite on Chrysalis with Intel and OpenMPI in optimized mode, using the current master as a baseline.

I have also run the remaining test groups that aren't part of the pr suite on their own.

Tests from test group run and are BFB (if there is a baseline comparison):

• GlobalConvergence
• GlobalOcean
• Hurricane
• IsomipPlus
• PlanarConvergence
• SphereTransport
• Tides

Manually changing resources before calling compass run has the desired effect:

• GlobalConvergence
• GlobalOcean
• Hurricane
• IsomipPlus
• PlanarConvergence
• SphereTransport
• Tides

[xylar]

@vanroekel and @sbrus89, this should address the issue with the Tides test group discussed in #456 (comment)

[xylar]

@sbrus89 and @vanroekel, I'm not quite done testing Hurricane and Tides because they haven't run to completion but I did verify that I could change the resources for both init and forward at runtime.

I would appreciate having you each test as well, though, just to make sure the issue if fixed.

[vanroekel]

I tested the hurricane and tides test case on chrysalis and confirmed a change to ntasks in the cfg file changed the number of cores used in the run. Thanks @xylar!

[xylar]

Seeing #482 but otherwise everything is working.

[xylar]

Thanks @vanroekel.

[xylar]

@sbrus89, I think @vanroekel's testing is sufficient. I'm going to merge.