V2: Add an instance for the new SeedGen type class

[lehins]

This is a slower, but more accurate alternative to #96

As mentioned in #96 this approach handles endianness correctly

[Shimuuar]

Lets go with this one. Mostly because it have less conditional compilation. Having support for both random 1.2 and 1.3 is painful enough. I don't want to support different versions of primitive as well

[Shimuuar]

I also played a with optimizing your implementation. It turns out -O2 has pretty big effect. With -O1 inner loops produced by stream fusion pass around tuples of int etc. I'll check whether it has any effect on rest of library and make release then