Prepare some benchmarks

[nadako]

I have some ideas about potentially improving performance of the library, for example, using macro-generated static Context instead of reflection-based one.

But one can't be sure it's worth it without proper benchmarks, so I should implement those first.

Also it would be interesting to compare performance between targets and also between different implementations such as mustache.js and original ruby mustache.