(** The following lemma (which one might expect, given how cofix in Coq
and guarded recursion in Iris work) should not be provable:
It requires us to conjure up a proof that that any acceptable expressions by inv
are in the simulation relation.
While we only get that after already having taken a step in source and target, thus justifying soundness, it still requires us to produce this proof without having seen the "full" co-inductive step, which should lead us to two expressions related by inv again.
If we fix the statement such that we only get the full coinduction hypothesis after having shown the full step "to the next iteration", we exactly arrive at the above statement [sim_lift_coind].
(but I still think this should be a sound co-induction principle: we'd just need a way "to look into the future" to justify it).