GEARRT: conservative flux exchanges

This MR does the same as !1411 (merged) did for Gizmo hydrodynamics: Change the RT interactions and flux exchanges to conservative ones.

Some additional attention was necessary for the RT sorts due to the sub-cycling.

This passes my tests, and I ran a couple of thousand (hundred) steps of EAGLE_12 (EAGLE_25) successfully (until something went wrong with hydro, which is a different problem.) Feel free to stress test this too if you so desire. There is a small catch though: The way to "chose" whether an RT scheme runs with symmetric MPI comms or not is a bit hidden. The MPI_SYMMETRIC_FORCE_INTERACTION macro is defined in src/part.h only when a flavor of Gizmo is selected as the hydrodynamics method. This remains unchanged: The RT will also only use symmetric MPI comms if a gizmo hydrodynamics flavor is selected.

As such, any RT scheme that runs on top of Gizmo hydrodynamics will also have symmetric flux exchanges. And vice versa: Any RT scheme running on top of SPH flavors will have SPH-like "force" loops.

@yuyttenhove , do you perhaps want to have a look at this MR too before the merge?

src/cell_unskip.c

@@ -3142,6 +3142,7 @@ int cell_unskip_rt_tasks(struct cell *c, struct scheduler *s,
            * on this node as well. */
           scheduler_activate_recv(s, ci->mpi.recv, task_subtype_rt_gradient);
           scheduler_activate_recv(s, ci->mpi.recv, task_subtype_rt_transport);
+          if (sub_cycle) cell_set_skip_rt_sort_flag_up(ci);
 #endif
         }
 
@@ -3188,6 +3189,7 @@ int cell_unskip_rt_tasks(struct cell *c, struct scheduler *s,
            * on this node as well. */
           scheduler_activate_recv(s, cj->mpi.recv, task_subtype_rt_gradient);
           scheduler_activate_recv(s, cj->mpi.recv, task_subtype_rt_transport);
+          if (sub_cycle) cell_set_skip_rt_sort_flag_up(cj);
 #endif
         }