This bug is strange:

- In order to type check the argument `I : gmap lft lft_names`, Coq needs an
  instance `Countable lft`. Since `finite.finite_countable` has been declared
  using `Hint Immediate` it will try that instance first (*), which means it
  has to find an instance `Finite lft`.
- Apparently, at this moment the type of the argument `A` has not been inferred
  by the type checker yet. So, there is a hypothesis `A : ?T`.
- In order to prove `Finite lft`, Coq unifies `?T` with `Finite lft` and uses
  the hypothesis `A`.
- Subsequent type checking fails, because `gmap lft lft_names` is type checked
  using a wrong/arbitrary instance `Countable lft`.

This is probably a Coq bug: type class search should not just unify with
hypotheses whose type is just an evar.

(*) `finite.finite_countable` has been declared using `Hint Immediate` to make
sure it is only used at leafs, not to make sure that it is used first.

Change the notation for borrows : now, it is just a notation for the predicate transformer (it does not include the payload of the borrow).
---
--- The new behavior is somewhat more intuitive, and better behaved wrt view predicates.

The inheritance of raw reborrowing is not (no longer) necessary. Let's remove it and call it shorten.

Get rid of down_close in creation.v : we can take all the lifetimes that are alive, which is simpler.

Instead, we strengthen a little bit the induction hypothesis.