Function safe_arch::reciprocal_sqrt_m256

#[must_use]pub fn reciprocal_sqrt_m256(a: m256) -> m256

This is supported with target feature avx only.

Reciprocal of f32 lanes.

let a = m256::from_array([16.0, 9.0, 4.0, 25.0, 16.0, 9.0, 4.0, 25.0]);
let b = reciprocal_sqrt_m256(a).to_array();
let expected = [0.25, 0.33333, 0.5, 0.2, 0.25, 0.33333, 0.5, 0.2];
for i in 0..8 {
  assert!((b[i] - expected[i]).abs() < 0.001);
}