XorShiftRng::new should use a non-constant seed

[brson]

It appears that all XorShiftRng's are initialized with the same seed by default. Looks like a foot-gun to me.

[brson]

I'm not entirely sure if it's ok to just seed with random numbers or not. Paper is here: http://www.jstatsoft.org/v08/i14/paper

[brson]

Apparently bad seeds make xorshift suck, so this may not be easy.

[omasanori]

FYI xorshift engines obviously work bad when w = x = y = z = 0.

[lilyball]

From skimming the paper, it suggests that the seeds can be drawn randomly from the set of integers, as long as they're not all 0. Looking at various results for Googling for "xorshift seed" seems to back this up.