Overly liberal with * in some type specifiers

[Douglas Katzman]

example 1.
(compile nil '(lambda (x) (the * x))) => #<FUNCTION>, NIL, NIL

example 2.
(compile nil '(lambda (f) (funcall (the (function (*) boolean) f) 1))) => #<FUNCTION>, NIL, NIL

but
(compile nil '(lambda (x) (declare (* x)) x)) => #<FUNCTION>, T, T ; correct answer

"*" should only be legal in compound FUNCTION specifier if ARG-TYPESPEC is omitted and VALUE-TYPESPEC is present in the metasyntax "function [arg-typespec [value-typespec]]"