Pysap-mri

Release with docs

- Added B0 correction
- Added WeightedSparseThreshold
- Added Cartesian 2D online MRI reconstruction

This release has some major bug fixes, some performance updates for FFT class, and added OSCAR `scale-based` regularization.

118 Refactor and remove unwanted imports in CircleCI
111 Fix FFT class, move to `scipy.fft` to better handle multi-process. Also fix the grid-search
113 Fix the normalize_frequency_locations code to work well with given `KMax`
114 Add oscar regularization based on `scale`

We have the following new features:

* Support for gpuNUFFT library : We can now get much faster reconstructions (In about 1.5 Hours for full 3D reconstruction for 44 channels)
* Support for density compensators estimation: This preconditions the k-space and drives towards faster convergence (Nearly 10-15min for the same reconstruction as above, with better performance)
* Reconstructors API for easy modeling of MR Image
* Iterative solvers like POGM, FISTA and condat-vu with support for metrics and cost functions to analyze convergence

Overall to use this, all you need to do is:
`pip install python-pysap`
If you have GPU with CUDA, you can use `pip install gpuNUFFT` for speed!