Additional tests for svmbir interface about scico

[Michael-T-McCann]

In case it is useful: there are comparisons already between the svmbir prox and a generic scipy solver in https://github.com/lanl/scico/blob/main/scico/test/linop/test_radon_svmbir.py (see test_prox and test_prox_weights).

[bwohlberg]

Good point. However, it looks as if those tests compare the svmbir prox with a prox computed using evaluations of the SVMBIRWeightedSquaredL2Loss loss function. We need additional tests to confirm that the svmbir prox also gives the same result as a prox computed using the svmbir forward and adjoint operators.

[bwohlberg]

ToDo:

• Add an initializer parameter (default None) for Functional.prox
• Add passing of appropriate initializer to prox in ADMM
• (and PGM?)
• Add tests based on operators and CG for prox of SVMBIRWeightedSquaredL2Loss
• Initializers for radon_svmbir.ParallelBeamProjector and SVMBIRWeightedSquaredL2Loss should set mask radius to sufficient size to avoid masking and provide parameters that allow setting of svmbir prox solver stopping tolerance and max steps
• Choose appropriate defaults for svmbir prox solver stopping tolerance and max steps

[Michael-T-McCann]

In branch mike/approx_prox proxes all take **kwargs and ADMM passes an initial value for the prox as v0.