The redundant baseline code is fairly inflexible.
It throws away all non-redundant baselines, which have use in sky based calibration.

Baselines are grouped in the uv-plane however there's no check whether baselines are conjugate

DFT based visibility code currently uses RegularGridInterpolator, Bella's work has shown that spline gives better accuracy

parallelisation causes multiple threads to write to the same array element.

The analytic covariance matrix matrix is off by a factor of 1.5

The covariance matrix is highly ill conditioned and the subsequently computed FIM as well.

Add calculation of cramer rao bound for real telescopes