Fix ys in Pollard's rho algorithm

Previously ys simply tracked the previous value of y. This doesn't make
a lot of sense because it means the final loop, meant to protect against
finding multiple factors at once, will only run for one iteration before
encountering the known non-1 value for G. In Brent's paper it's set
before the size-m loop instead of in its body. This is the more logical
location because the final loop is a backtracking step to go over the
last m or so values with the more accurage gcd instead of the faster
multiplication.

Author: mlochbaum

src/Primes.jl

@@ -411,8 +411,8 @@ function pollardfactors!(n::T, h::AbstractDict{K,Int}) where {T<:Integer,K<:Inte
             k::K = 0
             G = 1
             while k < r && G == 1
+                ys = y
                 for i in 1:(m > (r - k) ? (r - k) : m)
-                    ys = y
                     y = y^2 % n
                     y = (y + c) % n
                     q = (q * (x > y ? x - y : y - x)) % n