pub fn rms_norm( x: &[f32], w: &[f32], tokens: usize, dim: usize, eps: f32, ) -> Vec<f32>
2026-09-25: RMSNorm: w * (x * rsqrt(mean(x^2) + eps)) in f32, back to bf16.
w * (x * rsqrt(mean(x^2) + eps))