Reduce memory footprint of Lamino and USFFT

Purpose

This pull request addresses #66. Basically, the memory footprint of the Lamino operator is too large.

Approach

There are two approaches used in this pull request.

First, the checkerboard method for FFT shifting is implemented. This method uses properties of the FFT such that multiplying an array with a checkerboard of -1, 1 before and after the FFT is applied is equivalent to performing an FFT shift before and after.

Second, I dropped in the CuPy FFT functions that allow for the overwrite_x parameter which enables in-place FFT operations if available.

These two approaches have reduced the memory consumption of the Lamino operator by approximately half. Here is the memory usage before the changes:

tests/operators/test_lamino.py:49:test_adjoint (9.37GB, 5.57GB)
tests/operators/test_lamino.py:51:test_adjoint (8.06GB, 686.00MB)

Here is the memory usage after the changes:

tests/operators/test_lamino.py:49:test_adjoint (4.47GB, 3.42GB)
tests/operators/test_lamino.py:51:test_adjoint (3.87GB, 0.00B)

This is for n=16*14, ntheta=8*14, tilt=np.pi/3, eps=1e-3 with the Lamino adjoint test.

Pre-Merge Checklists

Submitter

[x] Write a helpfully descriptive pull request title.
[x] Organize changes into logically grouped commits with descriptive commit messages.
[x] Document all new functions.
[x] Write tests for new functions or explain why they are not needed.
[ ] Build the documentation successfully
[x] Use yapf to format python code.

Reviewer

[ ] Actually read all of the code.
[ ] Run the new code yourself.
[ ] Write a summary of the changes as you understand them.
[ ] Thank the submitter.

AdvancedPhotonSource / tike