Further schema refinement & beginnings of fixed-size specification

[skyreflectedinmirrors]

This MR adds the ability to:

• independently specify GPU / CPU overrides in schema (and test)
• apply memory limits in test matrix only to specific platforms (to combat NVIDIA's seeming inability to respect it's own memory bounds (re: https://devtalk.nvidia.com/default/topic/1019997/cuda-programming-and-performance/how-to-handle-cl_mem_object_allocation_failure-errors-if-amount-of-useable-memory-is-not-known-/)
• adds rudimentary control of the number of problems expected in the array-sizes. More work needs to be done to make this fully functional for wrapper / excecution standpoint (currently, we just issue a Critical warning about this). This will take place after the paper is submitted

Finally, this PR:

• unifies file naming performance / validation runs