Improve Argon2 benchmark for small memory

Merged username-removed-653601 requested to merge omos/cryptsetup:argon2-bench into wip-luks2

Hi,

I tried to play a bit with the Argon2 benchmark so it is not so slow on smaller memory cost limits (such as the default 1024 KiB). I believe it is now about 2 times faster for this case (20s -> 10s on my machine). It is still not perfect, but it should at least save us some console-staring time for now :)