Make benchmark output (stdout) more readable

[Flamefire]

This uses numpy to print the most important information about the benchmark run to stdout:

• script invoked
• scorep parameters
• summary of timings (number of repetitions, min, max, mean and median of times)

I moved the -m scorep out to scorep_settings so they are shown in the script summary. It also makes the code easier as only 1 param for enabling has to be passed: -m scorep and the settings instead of a bool and a potentially empty list which is ignored if that bool is false

Example output (part) looks like this:

#########
bm_simplefunc.py: ['-m', 'scorep', '--instrumenter-type=dummy']
#########
100000: Range=0.0000-0.0000 Mean=0.0000 Median=0.0000
200000: Range=0.0000-0.0000 Mean=0.0000 Median=0.0000
300000: Range=0.0000-0.0000 Mean=0.0000 Median=0.0000
400000: Range=0.0000-0.0000 Mean=0.0000 Median=0.0000
500000: Range=0.0000-0.0000 Mean=0.0000 Median=0.0000
#########
bm_simplefunc.py: []
#########
100000: Range=0.0000-0.0000 Mean=0.0000 Median=0.0000
200000: Range=0.0000-0.0000 Mean=0.0000 Median=0.0000
300000: Range=0.0000-0.0000 Mean=0.0000 Median=0.0000
400000: Range=0.0000-0.0000 Mean=0.0000 Median=0.0000
500000: Range=0.0000-0.0000 Mean=0.0000 Median=0.0000

(The zeros are because I commented out the actual invocation)
As you can see the output is directly usable for further analysis or at least provides a quick answer on how fast it is. This works well with #101 which allows to selectively run 1 or more test configurations

The width of the repetitions (first number) is automatically determined from the maximum value so all are right aligned for easier comparison. Example:

#########
bm_simplefunc.py: ['-m', 'scorep', '--instrumenter-type=dummy']
#########
 100000: Range=0.0000-0.0000 Mean=0.0000 Median=0.0000
 200000: Range=0.0000-0.0000 Mean=0.0000 Median=0.0000
 300000: Range=0.0000-0.0000 Mean=0.0000 Median=0.0000
 400000: Range=0.0000-0.0000 Mean=0.0000 Median=0.0000
5000000: Range=0.0000-0.0000 Mean=0.0000 Median=0.0000

[Flamefire]

CI failure is unrelated. Fix in #103

[AndreasGocht]

@rschoene Please have a look. I am ok with this.

[Flamefire]

Rebased to include the CI fix.

Can this be included soon? The change is basically cosmetical only (i.e. the created results.pkl doesn't change) and together with #101 is very useful in quickly evaluating the performance impact of a change (e.g. the CString change or the new maps from #105)

[rschoene]

Looks good to me