Change tfjs-wasm-simd iteration count to 40 (worst case count to 3)

This makes the runtime for the benchmark ~7s/9s/11s for jsc/v8/sm respectively on my M4 Max MBP.
1 file changed