Recommend mean((x, y)) rather than middle((x, y))

It seems more logical and simpler for users to recommend using the
same function. This also helps making Statistics a standalone package
(#128) as `mean` will be moved to Julia Base,
but `middle` will remain in Statistics.

src/Statistics.jl

@@ -109,7 +109,7 @@ if !isdefined(Base, :mean)
             f(itr)
         catch MethodError
             rethrow(ArgumentError("""mean(f, itr) requires a function and an iterable.
-                                    Perhaps you meant middle(x, y)?""",))
+                                    Perhaps you meant mean((x, y))?"""))
         end
         Base.reduce_first(+, f_value)/1
     end

test/runtests.jl

@@ -173,7 +173,7 @@ end
     @test isnan(@inferred mean(Iterators.filter(x -> true, Float64[])))
 
     # using a number as a "function"
-    @test_throws "ArgumentError: mean(f, itr) requires a function and an iterable.\nPerhaps you meant middle(x, y)" mean(1, 2)
+    @test_throws "ArgumentError: mean(f, itr) requires a function and an iterable.\nPerhaps you meant mean((x, y))" mean(1, 2)
     struct T <: Number
         x::Int
     end