NCCL support for Stacked op/array & Solver with doc and tutorial update

[tharittk]

This PR adds the completeness for NCCL support on operators (#137 ), solvers, and their corresponding tests. I've also updated the documentation to reflect these changes and added a write-up for a separated "Poststack Inversion" tutorial to demonstrate the migration from an MPI to an NCCL backend. The writing on the tutorial is considered 'drafted'

Change Made

• DistributedArray.py and cls_basic.py

  small changes are introduced to ensure base_comm_nccl of the input is passed through the output constructor.

• test_linearop_nccl.py, test_stackedlinearop_nccl.py, test_stackedarray_nccl.py, and test_solver_nccl.py

  These test files are similar to those in /tests but under NCCL environment. The tests involve complex number are commented out.

• Documentation and Tutorial (gpu.rst , poststack_nccl.py)

  Updated the documentation to add green checkmarks for the basic Ops, indicating full GPU support.
  "Poststack Inversion" Tutorial

Noted: the NCCL backend for subcommunicator, particularly asarray() introduced by @mrava87 in #139, is subjected to further debugging. This bug was not uncovered in previous test set.

[mrava87]

@tharittk good stuff!

I did leave some comments and some open thoughts on what to do with the nccl tutorials... feel free to suggest any other possible way to handle them 😄

[mrava87]

Very good work @tharittk, I am going to merge this 😄

[mrava87]

@tharittk I worry this may be broken in the online documentation as we don't build tutorials_poststack_nccl... I'll let it render and see, we can always make a small PR later to fix it :)