Added CGLS using DistributedArray

[rohanbabbar04]

• Added conj() and copy() to DistributedArray as they were used quite often.
• Added CGLS class similar to pylops but now using DistributedArray.
• Added cgls method which will be used by users, as we do in pylops(again with DistributedArray).
• Added some random tests for both real and complex(test cases are same as pylops).
• Added cgls and CGLS to docs.
• Also added cgls to __init__.py(same as pylops).

[mrava87]

This code looks very good. I just found something in the tests that I didn't like (the use of a very loose assert assert_array_almost_equal(.. decimal=4) and tried to switch to assert_allclose(..., rtol=1e-14) - I remember Yuxi asked something similar for the tests of DistributedArray.

I tried to make the change but the tests fail (some pass if I lift it to 1e-8)... this tells me something isn't the same as the pylops implementation. As long as we understand why this is the case, it is not a problem (perhaps MPI does some rounding when passing arrays?), but I would like you to investigate this a bit before merging and being happy with having some inconsistency with the serial solver

[rohanbabbar04]

So @mrava87 I was able to figure out as to why it was failing for rtol=1e-14, the main reason was the floating precision,
numpy dot product is a little imprecise so to handle this i updated float64 -> float128 and complex64 -> complex256 in the test cases and everything seems to work properly...

[mrava87]

Ok, interesting. Good finding :) then I think this is good to go