Optimize parsing by compiling regular expressions once

[omochi]

I have troubled with this library when I parse 100 user-agent strings at once.

Rule has regex source string and compiles it for every matching.
Compiling large regex such like used in this library is very heavy operation.

It changes Rule to have compiled regex rather than source string.
So if we test many strings, it can reuse compiled regex and improve performance.

In my case, in situation which takes over 6000msec originally,
it reduced to under 150msec.

[omochi]

Thanks for the merge :)

[malcommac]

Thank you for your optimization