problem sizes to benchmark

[ev-br]

Currently, we only run for an arbitrary pair of problem sizes. Running for all sizes from 100 to 1000 with a step of 1 is probably too much resource-wise, so some middle ground needs to be found.

[ev-br]

9911a94 runs multiple problem sizes. More fine-tuning of useful sizes is easy (see benchmarks.py) and should be done on a routine by routine basis.