[mcmehrtens]

Add opt-in singly degenerate mobility to the standalone Cahn–Hilliard integrator, so it can merge independently of agglomeration and Flame coupling.

Both SDCH methods use conservative face fluxes with arithmetic face mobility and a compatible discrete chemical-potential Laplacian. The spectral method applies an FFT-based discrete biharmonic stabilizer, including Nyquist modes. Constant mobility remains the default; its existing realspace and spectral update functions are unchanged.

Port the reusable Cahn–Hilliard additions from agglom-ddch-spectral:

• Virtual inheritance, protected state, configurable input/field names, and a spatial mobility-mask hook for later derived integrators. Masks are filled on native AMR layouts before spectral assembly, and realspace mask ghosts are filled across coarse/fine boundaries.
• Optional SDCH diagnostics: mobility, mask, cell-centered face-flux averages, conservative RHS, raw update, and actual field increment.
• Optional clipping, disabled by default to preserve conservation, and the existing branch's opt-in realspace timestep switch.
• Null-initialized owned pointers, expression initial conditions, mobility floors, and parameter validation.

Reduce the regression matrix from 38 sections to 12: 2D/3D analytic Fourier updates and checkerboard decay, mobility floor and stationary phases, strong-mobility dissipation, and partial spectral AMR coverage. Fold custom names and numerical diagnostic checks into an existing case. The checker reads native and HDF5 plotfiles; the test runner can select HDF5 binaries with --hdf5. Keep FFT CI coverage in both dimensions.

Validation:

• Built 2D and 3D with --fft --hdf5 --offline in each of alamo-llvm, alamo-gcc, and alamo-apple-clang, using the fish helpers and separate compiler build trees.
• In every environment, passed all 12 SDCH regressions (nine 2D and three 3D) plus the serial unit suite in both dimensions: 36 verified SDCH runs and six successful unit-suite invocations, with zero failures or skips.
• Passed two-rank HDF5 constant-mobility CH smoke runs in both realspace and spectral modes in all three environments. The existing native-plotfile-only CH regression checker was excluded from these HDF5 runs.
• Verified SDCH diagnostics and active clipping with HDF5 readback in Apple Clang and GCC: initial field 1.2, final field 1.0, raw update 1.2, increment -0.2, mobility zero, and mask one.
• Passed a two-rank LLVM realspace SDCH partial-AMR run with diagnostics and finite field readback on both levels. The spectral AMR regression is included in each environment's suite.
• Used uv with the latest installed Python, 3.15.1, for Python validation. Checked source equality across build trees, unchanged legacy constant-mobility kernels, and git diff --check.

SDCH requires periodic boundaries. Clipping can change the mean. Realspace AMR does not reflux coarse/fine fluxes. Spectral AMR assembles a full finest-level grid, with corresponding memory costs. These tests do not establish arbitrary-timestep energy dissipation, AMR convergence/restart accuracy, or MPI scaling. Flame/agglomeration equations, masks, restoration, and simulation inputs remain outside this PR.