However, there's still a bad admit left in iris_core...

introduce "CMRAExt", and extension property on CMRAs. We will need it to show that later commutes with star. For now, just assume it holds.

Conflicts:
	coq-ho/iris_ht_rules.v

On branch hackgreement
	modified:   coq-ho/iris_core.v
no changes added to commit (use "git add" and/or "git commit -a")

define view shifts and weakest-pre, and show they are all the things though ought to be (downclosed, non-expansive, monotone)

On branch hackgreement
	modified:   coq-ho/iris_plog.v
no changes added to commit (use "git add" and/or "git commit -a")

change SPred to be "bounded": they must hold a step-index 0. Change SPred-n-equality to require equivalence even at level n.

This gives rise to a nice lemma relation validity at level n, and n-equality. Plus, it works nicely for all existing constructions in lib/ (mainly because equality was already bounded)

The old
	fill_value K e : is_value (fill K e) -> K = empty_ctx.
won't work.

Counterexamples: K=(v,•) or K=inl • satisfy is_value K[v]
for reasonable choices of is_value.

reduce extensionality of fdFold to a commutativity lemma about fold_right (on lists) and permutations (of lists)

srengthen the induction principle and show euqalities such that the characterization of fdFold can be shown using fdRect

We had assumed fill left-injective. This doesn't hold for complicated
languages like the λ-calculus (counter-example: K = v • and K' = • v
so that K[v] = K'[v] but K ≠ K').

We only used the offending axiom, fill_inj1, to prove reasonable-looking
properties of context composition. Those are now axioms, and fill_inj1
is now gone.

Aside: For the λ-calculus, it's possible to prove (fill K) =1 (fill K') -> K = K'
where =1 is extensional equality. This does not seem strong enough
to prove the properties of context composition we want.