IB moment of inertia from the cell sum has an extra 1/cell_volume factor #1994

s_compute_moment_of_inertia (src/simulation/m_ibm.fpp) computes the moment numerically for geometries without an analytic formula (everything except circle, rectangle, ellipse, sphere):

moment = moment + distance_to_axis          ! sum of r^2 over the body's cells
...
moment = moment*patch%mass/(count*cell_volume)

patch%mass is the body's total mass (it is used as force/mass in m_time_steppers.fpp:908 and accel_bf*mass in m_ibm.fpp:1426). For count uniform-density cells each carries mass/count, so the moment is mass * sum(r^2) / count, with units of mass*length^2. Dividing additionally by cell_volume makes it mass/length, so the result scales with grid resolution (e.g. refining by 2x in 3D multiplies the moment by 8) and the angular response of two-way-coupled bodies (STL/airfoil/other non-analytic shapes) depends on the mesh.

Expected: moment = moment*patch%mass/count, which is consistent with the analytic branches (e.g. 0.4*mass*r^2 for the sphere). The line dates from #1075.

Separately, the sum runs over the local rank's cells only, with no reduction across ranks (the routine is a GPU_ROUTINE(seq) called from device loops, so an MPI reduction cannot simply be added there); a body split across ranks gets a different moment on each rank. #1992 fixes the periodic-marker decoding and the 0/0 on ranks holding none of the body, but not this.