Clean up the Smaps estimation module and add an easier way to use it through NonCartesianFFT directly

[chaithyagr]

We should also ideally add support for:

• [ ] Windowed Low frequency: Here we just do a density compensated adjoint of the gridded k-space data which is windowed using hamming or hanning window implemented by @paquiteau
• [ ] A simple radial thresholding function, current implementation thresholds on a square (rect function in k-space) which could lead to bizzare ringing artifacts in estimated Smaps in image space (as rect translates to sinc in image space)