performance improvements of [SD]DOT on A64FX.

[iha-taisei]

The current [SD]DOT kernel uses 2x loop unrolling.
There is room for optimization on A64FX, increasing this factor should improve performance.
I will prepare and submit a patch with this optimization.