handle running non-GPU jobs on GPU partitions

[smoors]

in some clusters/partitions, when submitting job in a GPU partition, you are required to request at least one GPU.
in our test suite we generate tests for running non-GPU jobs on GPU partitions that have the 'cpu' feature defined.
for those tests a GPU should not be requested unless required by the rules of the cluster.

currently, our tests act as follows:

• the gromacs and tensorflow tests do not request any GPU for such jobs, which will fail in clusters/partitions that require it
• the osu test, on the other hand, always requests one or more GPUs for such tests, depending on the number of cores requested.

as this is a cluster/partition-specific policy, ideally this would be a partition-specific setting in the site config file.
for example, we could add a feature must_request_gpus to the partitions that require it,
and add a check in the hooks for the presence of this feature via self.current_partition.features.