add Conjugate Gradient with Momentum

[Lenoush]

In base.py , add a first draft of cg function with
cg(self, kspace_data, x_init=None, num_iterations=10, tol=1e-4)
where x_init = initial image , tol = tolerance

Classique example of gradient descent :

def cg(fourier_op, kspace_data, num_iterations):
    L = fourier_op.get_lipschitz_cst()
    image = np.zeros(fourier_op.shape)
    for i in range(num_iterations):
         grad = fourier_op.data_consistency(image, kspace_data)
         image = image - grad * L
    return image

Help of :

• https://sigpy.readthedocs.io/en/latest/_modules/sigpy/alg.html#ConjugateGradient
• https://aquaulb.github.io/book_solving_pde_mooc/solving_pde_mooc/notebooks/05_IterativeMethods/05_02_Conjugate_Gradient.html#

[paquiteau]

I think this would be more suited in mrinufft.extras

[chaithyagr]

@Lenoush I think I agree with @paquiteau, can we move this to extras module. Also it would help to have test and an example?

[paquiteau]

This looks very good !
The example deserves more explanation (add a ref to the CG algorithm, Wikipedia is okay)

[paquiteau]

LGTM