Feat: proximal module

[mrava87]

This PR is a starting point of an effort to support distributed proximal operators and solvers from pyproximal in pylops-mpi

To begin with the following operators/solvers are supported:

• any separable proximal operator (i.e., where prox is taken element-wise) via simple wrapping on the pyproximal operator into a MPIProxOperator
• A custom distribured proximal operator for L2 norm named MPIL2 which leverages distributed stacking and solvers to solve augumented systems of equations in the prox operator.
• ProximalGradient and ADMML2 solvers

These already enable the solution of 2 typical use-cases of convex optimization:

• sparsity promoting inversion: added to reflectivity tutorial
• TV-reg inversion: added to poststack tutorial - note that for this to be possible, I needed to extend StackedDistributedArray to operate with nested StackedDistributedArray. This will be handled separately in Feat: allow nested stacked arrays #201 and then merged back into here.

However, it is intentional that the operators/solvers implemented here do not always match those in pyproximal with some features currently missing, namely:

• ProximalOperator, some advanced features - Advanced features in MPIProxOperator #205
• ProximalGradient, some advanced features - Advanced features in ProximalGradient #204

TODO:

• Improve description of separability in MPIProxOperator docstrings
• Add nccl proxsolver tests
• Add sparse inversion to reflectivity GPU tutorials

[mrava87]

@rohanbabbar04 I have worked for quite a bit on this new (pretty big) feature... basically I am aiming to add a module called proximal which will implement a distributed version of pyproximal. However, I will be taking it bit by bit, and this is the first PR with minimal components required to be already able to solve a subset of the problem we tackle in pyproximal end-to-end.

It would be great if you could go through and provide your review 😄

[rohanbabbar04]

Thanks @mrava87, I will take a look at this in a couple of days.

[mrava87]

Btw I noticed something strange lately.. in some PRs sometimes one action fails just at the start of the tests... but then if I retrigger them they eventually succeed... an example just occurred in this PR