Verify the results and split up the benchmarks

[lemire]

Verify the results and make sure we are comparing apples to apples. In particular, we should make sure that the conversion is correct and that the strings produced are similar. We could have different benchmarks depending on the precision of the conversion. Shortest length is one type of precision, and it can't be directly compared against fixed precision.

[jaja360]

With #14, #27, #30 and now #36 which allows to evaluate properties of algorithms on any subset of values, is there something else we should add before closing this issue ?

I can separate the array like the one in algorithms.h:initArgs, such that we have instead two arrays: one of algorithms which in theory support shortest length (all except std::to_string, snprintf and abseil) and another one of algorithms which support fixed precision (some algorithms support both and would be in both arrays). Should we do that ?

[jaja360]

I don't think there's anything left to do for this issue. I'll close it.