GER kernel notes

As in GEMV, GER blocks $X$ so that it is reused in L1. Since the dominant direction of the loop is expected to be down the columns of $A$, incY is not restricted to 1, while incX is. As in GEMV, all routines except SYR2 block only the the $X$ dimension, while SYR2 blocks both, so that $A$ can be reused in L1.