Error bars for runtime profiles #68
Description
Following up on issue 1019 in the superseded coco repository: for each problem (like 24x51 in the default aggregation over all functions and 51 for each function), we could annotate the median runtime in the profile with percentile bars. This is different from constructing an entire new runtime profile (though would use the same data as a runtime profile of some percentile as suggested in the last comment of the above issue but plotted in a different order and at different y-values). When all runs were successful, this is also straight forward to understand and undoubtedly "correct". We could see (occasionally) very different bar widths for similar or even the same runtimes which may look quite erratic and difficult to read.