Enable one to import both stdpp's base and ssrfun.

Note that (f x.1) now parses as (f (fst x)) rather than (fst (f x)).
(This change affects one proof in Iris.)

This gets rid of the old hack to have specific notations for pairs
up to a fixed arity, and moreover allows to do fancy things like:

```
Record test := Test { t1 : nat; t2 : nat }.

Definition foo (x : option test) : option nat :=
  ''(Test a1 a2) ← x;
  Some a1.
```

This one works for setoid rewriting under binders.

- Name all variables that we refer to.
- Put types in definitions.

This follows the associativity in Haskell. So, something like

  f <$> g <$> h

Is now parsed as:

  (f <$> g) <$> h

Since the functor is a generalized form of function application, this also now
also corresponds with the associativity of function application, which is also
left associative.

This is similar to `f_equal/=`.

Also make the instances non-global, to prevent multiple instance
problems.

We prove that various types are infinite, notably:
- nat, N, positive and Z;
- string (using pretty-printing of nat);
- option, with an infinite element type;
- list, with an inhabited element type.

Furthermore, we instantiate Fresh for strings.

This implements a simple linear search for fresh elements.

This generalizes Fix_unfold to a setoid setting. In particular,
we can use this to unfold multi-argument fixpoints without
requiring functional extensionality.

Infinity is described by having an injection from nat.

The documentation for some typeclasses used the wrong
names for these typeclasses.

This addresses some concerns in !5.

This way, we will be compabile with Iris's heap_lang, which puts ;;
at level 100.