Consolidate benchmark generating programs

[xrchz]

Currently there are two similar (but not identical) Python programs compiler/benchmarks/cakeml_benchmarks/benchmark.py and .../mlton_benchmarks/benchmark.py. This issue is to refactor these programs so that only 1 or fewer Python programs are required.

For example, this might involve moving the differences between the files above into a configuration file or command line arguments.

Ideally, the Python functionality would be removed altogether (e.g., replaced by a CakeML program), or reduced to a program with a clear function and interface (e.g., just for creating a plot given a data file and configuration arguments).

[ordinarymath]

Note that currently the benchmark is currently broken in multiple ways. Subsumed by #1126