1. There was a bug. When we approximate, for example 10 with power of
two - we got 16 (2^4). So register size must be 4. But instead register
if size 5 was used. It's not efficient.

2. Even we have efficient std::distance and operator+(int n) functions
dumb_advanced and dumb_size doesn't uses them, so they was replaced with
std::advance and std::distance accordingly

…hoxnox-shuffled

A bunch of other minor changes.