Reimplement MPCD for HOOMD 4

[mphoward]

Description

This PR finalizes the reimplementation of MPCD for HOOMD 4. The changes added to this branch that haven't been in a previous PR are:

• Add Yashraj Wani (@YMWani) to the credits for giving me the code that refactored the virtual fillers.
• Update default for BUILD_MPCD to match BUILD_MD and document hoomd.mpcd in BUILDING.rst.

See meta issue #775 and PRs #1580, #1655, #1681, #1689, #1690, #1691, #1712, #1714, #1734, #1737, #1741, #1763, #1778 for additional information.

Motivation and context

Resolves #775

How has this been tested?

All PRs have been tested along the way with C++ and Python tests. I additionally ran some applications with the new code and compared to HOOMD 2.9.7. The tests were: planar pressure-driven (Poiseuille) flow, planar shear (Couette) flow, and bulk diffusion of monomers. The figures below show identical results with the two versions of HOOMD.

Change log

* ``hoomd.mpcd`` reimplements the MPCD method for simulating hydrodynamic interactions in HOOMD 4. See the `migrating` page for an overview and individual class and method documentation for more information.

Checklist:

• I have reviewed the Contributor Guidelines.
• I agree with the terms of the HOOMD-blue Contributor Agreement.
• My name is on the list of contributors (sphinx-doc/credits.rst) in the pull request source branch.

[joaander]

Thanks!

I found and removed a few remaining references to GPU polymorphism.

I was unable to build this branch with HIP. Many mpcd files include cuda_runtime.h, reference cudaSucess / cudaError_t, and possibly make other references to CUDA APIs I didn't notice yet. @mphoward Would you like me to proceed with merging this now and making mpcd available for CUDA only in the next release? I'm OK with fixing HIP support for MPCD in a future pull request.

[mphoward]

Thanks! Yes, you’re right, and I didn’t notice this. Jens probably did not port any of the MPCD code to HIP because it would not compile with polymorphism. If we still have to fix these calls, it would be best to do it with hipper rather than HIP, so I agree with merging this now (with CUDA support only) for the next release & fixing it later!