Self-gravity is frozen at the old-time density during SDC sweeps

Locations: Source/driver/Castro_advance_sdc.cpp:94-96,
Source/gravity/Castro_gravity.cpp:45-53,125,300-328, and
Source/gravity/Gravity.cpp:3476-3486.

Finding

Each SDC node stages its current solution in S_new/Sborder, but calls
construct_old_gravity(prev_time). The gravity source then reads
get_old_data(Gravity_Type), not a gravity field solved from the current node's
density.

For Poisson gravity, construct_old_gravity explicitly skips the old solve in
single-level runs. True SDC rejects AMR, so all supported Poisson/true-SDC runs
follow this skip. Repeated SDC sweeps therefore do not update the gravitational
field as the density changes within the step.

The integrated system uses frozen g(t_n), rather than g(rho_node). This loses
the intended temporal order for evolving self-gravitating flows. Constant
prescribed gravity is not affected by this particular defect.

Evidence

Static inspection of the stage driver, old-gravity constructor, source lookup,
and gravity RHS selection confirms that no stage Poisson solve occurs. The
driver already has a TODO acknowledging that it is not using the current stage
density. This report does not claim that post-timestep gravity solves are
absent; such solves cannot correct the preceding stage evaluations.

No self-gravity convergence run was performed.

Proposed fix

Introduce a stage-gravity evaluation that solves from the staged node density
and passes its acceleration explicitly to the source builder. Keep physical
old/new gravity state and metadata intact rather than overloading old-time
fields. Refresh the final gravity state consistently after the last update. If
this cannot be supported yet, explicitly reject evolving self-gravity with true
SDC rather than silently claiming its formal order.

Validation after fixing

Use a smooth single-level self-gravitating problem with changing density.
Compare temporal convergence at fixed fine spatial resolution for SDC2 and SDC4.
Check that node acceleration changes within the step and that constant-gravity
regressions remain unchanged.

Start with the listed locations in Source/driver/Castro_advance_sdc.cpp, Source/gravity/Castro_gravity.cpp, and Source/gravity/Gravity.cpp; trace how staged SDC density reaches gravity source construction and old-data lookup. Use the proposed stage-gravity evaluation and explicit acceleration flow as the design target. Done means node acceleration changes with density, SDC2 and SDC4 show temporal convergence on a smooth single-level self-gravitating problem, and constant-gravity regressions remain unchanged.