fix[next-dace]: Update connectivities in fastcall

[edopao]

Update the connectivity arrays in SDFG fastcall. In baseline, the connectivity arrays were only set on the first call and then were not updated, because they were assumed to not change during application lifetime. Although this assumption is true in general, it does not hold in some specific cases. During MPI tests in ICON4Py (#692), for example, it was observed that connectivity arrays get reallocated during a test session, because the default scope of some fixtures is function-level. For this reason, and also to align the runtime behavior of DaCe backend with GTFN, this PR always sets the connectivity arrays at each program call.

[philip-paul-mueller]

I think it looks good I have some very minor comments, nothing blocking.
However, I still think that the assumption "everything is allocated once" is a good assumption especially for the grid.
So I would suggest to add some configuration option like STATIC_ALLOCATED_GRIDS or GRID_LIFETIME=global.
However, I am also fine with a TODO.

[edopao]

I think it looks good I have some very minor comments, nothing blocking. However, I still think that the assumption "everything is allocated once" is a good assumption especially for the grid. So I would suggest to add some configuration option like STATIC_ALLOCATED_GRIDS or GRID_LIFETIME=global. However, I am also fine with a TODO.

Thanks for the review. I agree about being able to configure the behavior, I have added a config variable.

[edopao]

The AMD CI-pipeline fails because of vCluster issues.

[philip-paul-mueller]

LGTM