More benchmarks and longer benchmarks

[sbryngelson]

• More benchmarks are needed (3D cases that are homogeneous in 2 of the dimensions might be easiest to setup).
• Longer benchmarks are needed (more time steps, e.g.) for GPU cases, which run quickly and are thus vulnerable to small runtime differences. This can be passed into the case.py benchmark case via an argument that contains the device type, like examples/weak_scaling.

[wilfonba]

This is on my ToDo list.