Add option to run benchmarks until a certain precision is reached.

For some benchmarks I currently use `-L` to get accurate measurements.

In a perfect world I would rather specify run until you are 95% confident that the measured time is within +/-X% of the given value. Where I can specify both the confidence Intervall and the range.

No files, tests, or entry points are named in the issue. Begin by locating how benchmark duration and the -L option are handled, then clarify the statistical definition of confidence and range; done means an option can stop runs when the requested precision is reached.

Written by the indexing model from the issue text.