cl-preloaded.el: Fix the type lattice

We generally want types to form not just a DAG but a lattice.
If objects can be both `keyword` and `symbol-with-pos`, this
means there should be a more precise type describing this intersection.
If we ever find the need for such a refinement, we could add
such a `keyword-with-pos` type, but here I took the simpler
route of treating `keyword` not as a proper built-in type but
as a second-class type like `natnum`.

While fixing this problem, also fix the problem we had where
`functionp` was not quite adequate to characterize objects of type
`function`, by introducing a new predicate `cl-functionp` for that.

* lisp/emacs-lisp/cl-preloaded.el (cl-functionp): New function.
(function): Use it.
(keyword): Don't declare it as a built-in type.
(user-ptrp): Remove redundant declaration.

* lisp/emacs-lisp/cl-generic.el (cl--generic--unreachable-types):
Delete constant.
(cl-generic-generalizers): Remove corresponding test.

* lisp/emacs-lisp/cl-macs.el (cl-deftype-satisfies): Add entry for
`keyword` type.

* lisp/emacs-lisp/comp.el (comp-known-predicates): Fix type for
negative result of `characterp`.  Remove duplicate `numberp` entry.
Fix types for `keywordp` now that `keyword` is not a built-in type any more.

* test/src/data-tests.el (data-tests--cl-type-of): Add a few cases.
Remove workaround for `function`.

lisp/emacs-lisp/cl-macs.el

@@ -3467,6 +3467,7 @@ Of course, we really can't know that for sure, so it's just a heuristic."
                '((base-char	. characterp) ;Could be subtype of `fixnum'.
                  (character	. natnump)    ;Could be subtype of `fixnum'.
                  (command	. commandp)   ;Subtype of closure & subr.
+                 (keyword	. keywordp)   ;Would need `keyword-with-pos`.
                  (natnum	. natnump)    ;Subtype of fixnum & bignum.
                  (real		. numberp)    ;Not clear where it would fit.
                  ))