We define a stronger rule `related_bind_up`, in which there is a baked
in semantic type `R`. The idea here is that we don't actually require
the expressions that we bind to have the same syntactic type.
```
{E;R::Δ;⤉Γ} ⊨ e1 ≤log≤ e2 : τ
∗ (∀ vv, ⟦ τ ⟧ (R::Δ) vv -∗ {E;Δ;Γ} ⊨ K[v1] ≤log≤ K'[v2] : τ')
____________________________________________________________
{E;Δ;Γ} ⊨ K[e1] ≤log≤ K'[e2] : τ'
```
We can then use `bin_log_related_weaken_2` to prove the original
binding rule.
The advantages of the new rule is that it allows us to prove the
following compatibility rule for seq:
```
{E;(R::Δ);⤉Γ} ⊨ e1 ≤log≤ e1' : τ1 -∗
{E;Δ;Γ} ⊨ e2 ≤log≤ e2' : τ2 -∗
{E;Δ;Γ} ⊨ (e1;; e2) ≤log≤ (e1';; e2') : τ2.
```
The idea here is that we can also pick any *semantic* type to related
e1 and e1'. For instance, if both e1 and e1' are expressions of type
Nat then it is not necessarily the case that we can relate them at
that type -- they might reduce to two different numerals -- but
it *should* be the case that we can relate their effects, if it makes
sense. E.g.
((#l <- #1;; #0) ;; e) ≤ ((#l <- #1;; #1) ;; e)