fn data_fit(axm: &[Complex64], y: &[Complex64], m: usize) -> f64
Plain data fit ||A x - y||^2 / m, the quantity the divergence guard watches.
||A x - y||^2 / m