Lamino crash hotfix

[dyang37]

This is a hotfix to the laminography crash issue. This PR provides a hook for a more fundamental fix in the future.

As discussed, we add an additional parameter "lamino_mode" to cone3D.recon(). This parameter is automatically set by recon_lamino(), such that a more conservative parallel update pattern is used for laminography reconstruction.

docstring:

Test

• Cone beam recon: tested demo_3D_shepp_logan.py to make sure cone beam reconstruction is not affected.
• Brendt's test case: the case is reproduced with test/test_lamino_crash.py. The hot-fix works on a cluster node with 24 CPU cores, but does not work on a cluster node with 128 CPU cores.

In the future, we might need to make N_G a function of #CPU cores.

[cabouman]

It seems to run fine. So I am going to accept the PR.