Refactor ConstantAlpha to allow running with turbulence model

[ewquon]

Previous approach:

• Calculate shear stress (tau), which in general is strain rate multiplied by 2*(mu + mu_t) -- here mu and mu_t refer to the molecular and turbulent viscosities
• When calculating the diffusion contribution, multiply by rho/rho0 if using ConstantAlpha -- here, rho is the instantaneous local value, averaged from cell centers to faces
• The issue arises because mu_t also ends up being scaled by this density ratio

New approach

• Calculate shear stress with appropriate mu_eff, based on whether we're using the Constant or ConstantAlpha solver path -- here, tau is scaled by the cell-centered or face-averaged value of rho, depending on the stress quantity
• Diffusion contribution calculations are agnostic to the molec_diff_type
• Because the density used to scale the diffusion contribution is different, the gold data for regtests with ConstantAlpha will change

Tests based on DensityCurrent

• ConstantAlpha with les_type=none: dynamicViscosity=75, rho0_trans=1 (default) vs dynamicViscosity=150, rho0_trans=2 -- verified results were identical
• ConstantAlpha with les_type=Smagorinsky: dynamicViscosity=75, rho0_trans=1 (default) vs dynamicViscosity=150, rho0_trans=2 -- verified results were identical
• very small differences observed with and without Smagorinsky in this case

[ewquon]

Density current prob (coarse dx=dz=100 m) at 900 s, before/after this PR

[AMLattanzi]

Fairly certain the cell_data must be averaged to the strain locations (as is done in DiffusionSrcForState_N/T.cpp. Once that is completed, I am good with this PR.

[AMLattanzi]

For the off-diagonal stresses, shouldn't the cell_data be averaged to the location of the strain (this would hold for other places in the code as well)?

[ewquon]

Yes, you're right. I have a fix coming.

[ewquon]

@AMLattanzi, 23d55a5 fixes the segfault that showed up in the CI and 1834373 does the averaging that's needed.

[ewquon]

Actually hold on, I only updated the terrain solver path.

[ewquon]

Ok, @AMLattanzi, this latest commit makes the rest of the changes. I've verified that the rho averaging for the ConstantAlpha path matches the original mu_turb averaging.