Switch to crypto/rand

[ChimeraCoder]

Since we're not using fixed seeds in tests, there's no reason no use math/rand: #7

[bobby-stripe]

crypto/rand is still significantly slower here than math/rand unfortunately. In #43 we moved to using a sync.Pool of *math/rand.Rand objects seeded from crypto/rand: https://github.com/stripe/goforit/blob/master/internal/safepool/rand.go#L11-L18 . Going to close this for now, as I think we're still years away from Linux's getrandom improvements being deployed widely enough to directly use crypto/rand here.