peakflops() performance

[ViralBShah]

On Apple M2 Max, I have peakflops on OpenBLAS 0.3.29 giving me 3.5e11, whereas BLIS 2.0 is giving me 5.3e10.

cc @imciner2 @giordano

[imciner2]

I am not surprised too much by a difference here. BLIS probably hasn't had as much optimization done to it for the newer architectures compared to OpenBLAS considering its contributor base tends to be smaller.

[ViralBShah]

Running with BLAS.set_num_threads(1) I get the same flop rate on both.