Increment the itoa benchmark versions.

[Alexhuszagh]

Closes #144
Closes #148

[Alexhuszagh]

I've added scripts to auto-generate the benchmarks from various commits and plot them and then we can update the references just by changing the commit ID. It shows on average itoa outperforms lexical for integer writing, while lexical outperforms ryu (using dragonbox) slightly for float writing, we slightly outperform the core library for integer and float parsing.