Remove prefetches from SVE kernels

This is a precursor to enabling the SVE kernels for Arm(R) Neoverse(TM)
V1 which has 256-bit SVE. Testing revealed that the SVE kernel was
actually worse in some cases than the existing kernel which seemed odd -
removing these prefetches the underlying architecture seems to do a better job
:smile_cat:

Author: Mousius