Two aligned wind turbine example case

[ia267]

Created the flow case for the wind turbine example, including an interface for the turbine models (currently only the Actuator Disc Model)

changes

• turbine_model_t - abstract interface for turbine forcing models, with methods init, update, project_forces, write_output. Ships with a no-op dummy (turbine_dummy_t, also the iturbine=0 model) so the case runs before any real model exists.
• wind_turbine case - uniform inflow / convective outflow in x, using a polymorphic turbine_model_t. The case owns the turbine object and is the only place the iturbine switch appears (at allocation time); after that it calls the polymorphic turbine with no branching.
• iSVV / hyperviscous dissipation - added c_nu and nu0_nu for the compact6-hyperviscous scheme. This is the implicit LES the two_turbines case uses ( Incompact3d jles=4) - implicit numerical dissipation, not an explicit SGS model (the explicit Smagorinsky LES issue is Add Smagorinsky LES with wall damping #321). These params are optional.
• Boundary conditions - Dirichlet / periodic / periodic (matches the cylinder case). In Incompact3d the example uses free-slip walls in y and z, but the FFT Poisson solver currently requires z periodic and supports only a single MPI rank in non-periodic directions.
• Actuator Disc Model (ADM) (src/turbine/adm.f90) - super-Gaussian force smearing (King et al. 2017), local (Calaf) thrust coefficient C_T' = C_T/(1-a)^2, optional T_relax velocity filter, per-disc disc<N>.adm diagnostics, and restart-safe checkpoint state. This is a close representation of Incompact3d's ADM. The projected force divided by rho_air so it's a true acceleration (identical to Incompact3d when rho_air=1). Read from a .ad disc-coordinates file

Other changes

Not part of the disc model, but needed for the case.

• Cylinder case - MPI correctness fix. cylinder.f90 averaged the inflow/outflow mass flux over the local ny*nz (with a NOTE admitting this was wrong under y/z decomposition). Now uses the global plane size via get_global_dims. Unchanged on a single-rank, only fixes decomposed runs
• Q-criterion formula fix. compute_qcriterion (CUDA + OMP) now uses the full second invariant; the previous form dropped the 0.5*(div u)^2 term. Zero for divergence-free flow, but more accurate on the discerete field
• Backend x-face primitive + slice_max_sum min. The outflow BC reproduces Incompact3d's outflow + pre_correc, which needs two new backend capabilities. slice_max_sum gains an optional signed min_val (for the convective speed cx = 0.5*(uxmax + uxmin) * gdt/dx), and a new field_add_const_x_face (abstract + CUDA + OMP) for the exact outlet mass correction (bxxn = bxxn - ut + ut1). Without the mass correction, an under-resolved wake reaching the outflow injects boundary divergence; with it the example holds machine-precision divergence at the outlet. This is used for the wind turbine case's boundary treatment

Supporting cases

• inlet-noise BC: field based Dirichlet inlet refreshed per subsetp only when inlet_noise /= 0; RNG seeded deterministically (rank-offset) for reproducible runs at a fixed decomposition.
• base_case: finalise_case_specific hook, append -on-restart for monitoring.csv and central release of the persistent inflow-BC planes
• config: new wind_turbine_ml params (inlet_noise, adm_coords, rho_air, T_relax)
• docs: input_file.rst documents wind_turbine_nml, the ADM, the .ad format and the corrected Q-criterion

The example uses a dimensional (effective LES Re = 450) setup rather than Incompact3d's non-dimensional Re=45000, which is under-resolved at affordable cost and blows up; the dimensional formulation resolves the wake and runs cleanly to t=800.

Closes #309
Closes #311

Example case visualisations:
https://github.com/user-attachments/assets/f7b88c46-9b2a-43b8-9cc1-8921e9595fe2