Openmp cce

Co-authored-by: Daniel Vickers <danieljvickers@login11.frontier.olcf.ornl.gov>
Co-authored-by: Daniel Vickers <danieljvickers@login08.frontier.olcf.ornl.gov>

This PR is to expand the SG EoS to its general form.

Such a change is needed for the PC model to work on both 5 and 6 equation models

Including a suggestion from coderabbitai, effectively initializing qv_avg to input parameter on 's_compute_speed_of_sound'

…owCode#1274)

…and MPI communication

Features:
- Lagrangian bubble movement with projection-based void fraction smearing
- Kahan summation for accurate void fraction boundary conditions
- Extended MPI communication for moving EL bubbles
- New 2D and 3D moving Lagrangian bubble examples
- Updated test cases and golden files

This commit adds normal collision forces. The point is to checkpoint the code before merging with upstream. Collisions are not yet ready for use. Need to fix beta buffer filling first.

…e/moving-solid-particles

Buffer communication of beta is consistent with how the bubble solver fills beta.

There is an initial implementation of collisions, but this is not yet ready for use.

Various files that were different from master were updated to match.

…feature/moving-solid-particles

Total gaussian weight computation is now done in the particle dynamics loop

Removed WENO reconstruction of gaussian smeared particle quantities for gradient fields and using fornberg finite differences instead. WENO is overkill and a time sink.

Re-added fornberg finite difference weight computation.

Particle Solver:

1. Call rhs functions

1.1 Particle Dynamics subroutine
  GPU loop across all particles:
    - Compute fluid force on the particle
    - If two way coupled, spread particle force and volume fraction across surrounding cells with gaussian
    - If collisions are off, update particle velocities and acceleration

  - Finalize volume fraction field by filling buffers

  - If collisions are on, call collision subroutine to compute collision forces with nearby particles and with any solid boundaries. Then update particle velocities/acceleration
    - Collision subroutine:
    - Ghost particles temporarily added to arrays
    - GPU loop across all particles:
	- Check for overlap with neighbor cells, including ghost particles
        - Collision force for particle pair only computed once
        - Each particle gets checked for wall collision

    - Communicate forces applied to ghost particles

1.2 Particles Source Subroutine
  GPU loop across all domain cells:
    - Update fluid solver rhs with particle source terms for mass, momentum, energy

2. Call Particle RK update
   - Update particle positions/velocity
   - Transfer particles across ranks
   - Enforce boundary conditions on the particles. Some overlap is allowed with solid walls to allow the collision force to handle moving the particle away from the wall naturally.
   - Update particle volume fraction field (zero out the field variables, compute the gaussian contribution for normalization, apply the gaussian kernel, communicate beta field).

Optimization:

    - particle volume fraction and source term contributions at each cell are updated with an atomic update in the gaussian kernel function. Using particle cell linked lists makes this much slower.
    - Particle dynamics particle loop handles the fluid force and the gaussian kernel for projection all in one. This reduced the number of GPU kernels launched and reuses particle data already in memory.