protocols where I want to prove something called robust safety.
Ironically, to even state robust safety requires Hoare triples that
don't imply safety. So Iris supports both {P} e {Q} (implying safety)
and [P] e [Q] (not). I'll add a rule for forgetting about safety:

	{P} e {Q}
	— Unsafe
	[P] e [Q]

some time soon.

Aside: I'm an SSReflect weenie and know next to nothing about the
usual Coq tactics. My proof script changes likely reflect that fact.

write an interface to world_prop, and make iris_core (assertion logic + world satisfaction) work with it

proofs. One change to axiomatisation was needed.