This test takes a long time. It tests ca. 8900 conversion
combinations, and then it also runs each conversion for
at least 100ms in order to come up with some kind of benchmark.
Remove the benchmarking from the unit test, we have a separate
benchmarking tool for that now.
Also split the conversions into groups and run those as
separate checks, which allows better parallelisation at
the runner level (normal runs and when using valgrind).
3288093c75