While such activity parameter combination does not make sense, it is
technically possible to spawn activities with a group_id set but without a
grouping_method_id.
The consequence is zombie activities: they never fail, do not prevent other
activities from being executed, but and are permanently run on the node
which reserved more than one because:
- all but one activity is in MESSAGE_NOT_EXECUTED state, forcing an abort
- finalizeMessageExecution does not understand that MESSAGE_NOT_EXECUTED
  activities may not have failed, and itself fail while trying to detect
  the failure type (ConflictError vs. others), failing to free & postpone
  the activity, in turn failing to bring it to eventually permanent failure
  state.
Make those conditions consistent, and use tuple-unpacking to detect any
future inconsistency much earlier, simplifying debugging.