reduce overhead of println

[tathanhdinh]

This PR tries to reduce the (tiny) overhead that println! may make (because of locks and since stdout is line-buffered). I don't think that using a buffered writer over stdout will make a noticeable performance gain in this case, so feel free to reject it if you don't like.

[sharkdp]

I don't think that using a buffered writer over stdout will make a noticeable performance gain in this case

I also don't think so. Within hyperfine itself, I don't think we have to care about performance too much because a typical run of hyperfine will be completely dominated by the actual benchmarking time (which is on the order of seconds) whereas everything else (calculations, output) will run in a few milliseconds.

The println!-based code is a bit easier and clearer, so I'd like to keep it.

Anyway, thank you very much for your contribution 👍