fix[next-dace]: return no successors for terminal blocks in find_successor_state

[edopao]

Problem

find_successor_state (in gt4py.next.program_processors.runners.dace.transformations.utils) walks the outgoing region edges of a control-flow state, ascending into parent regions when a block has no successors locally. If the queried block and all its enclosing regions are terminal, the walk reached the SDFG root region and then recursed with the root region itself as state, so graph = state.parent_graph was None and the subsequent graph.nodes() call raised AttributeError: 'NoneType' object has no attribute 'nodes'.

This was hit via MultiStateGlobalSelfCopyElimination, which calls find_successor_state on transient-defining states (e.g. from OIR programs whose defining state is a terminal top-level block or inside a terminal LoopRegion).

Fix

Treat the root region as the termination point of the walk: the no-successors early-return condition in the inner _impl of find_successor_state now also triggers when graph.parent_graph is None (i.e. graph is the SDFG root region), returning an empty successor list instead of recursing into None.

Tests

New file tests/next_tests/unit_tests/program_processor_tests/runners_tests/dace_tests/transformation_tests/test_utils.py with two regression tests:

• test_find_successor_state: a terminal top-level state returns [] (previously crashed with AttributeError), a non-terminal state returns its successor.
• test_find_successor_state_terminal_loop_region: the last state of a LoopRegion body, where the loop itself is terminal at the top level, returns [] instead of crashing.

[philip-paul-mueller]

LGTM.

It looks good, but while we are at it, we should also add a unit test that handles a nested SDFG.

[philip-paul-mueller]

LGTM