Fix incorrect guard type refinement with length/map_size and != operator

[lukaszsamson]

Previously empty_list() and empty_map were returned for length(x) != n and map_size(x) != n when n > 0. Now the types are widened to list() and open_map(). List/map of size exactly n when n > 0 cannot be expressed in the type system. The problem does not affect tuples as their sizes are encoded in types

Fixes #15326

[josevalim]

This cannot be an empty list, right?

[gldubc]

This is a regression: they weren't tested explicitly previously, but things like length(x) == 1 should rightfully deduce for x the type of a non empty list.

These tests should succeed:

assert typecheck!([x], length(x) == 1, x) == dynamic(non_empty_list(term()))
assert typecheck!([x], not (length(x) != 1), x) == dynamic(non_empty_list(term()))

assert typecheck!([x], 1 == length(x), x) == dynamic(non_empty_list(term()))

assert typecheck!([x], map_size(x) == 1, x) ==
               dynamic(open_map() |> difference(empty_map()))
assert typecheck!([x], not (map_size(x) != 1), x) ==  dynamic(open_map() |> difference(empty_map()))

assert typecheck!([x], 1 == map_size(x), x) ==  dynamic(open_map() |> difference(empty_map()))

[josevalim]

Yes, we need a few additional tests and cases!

[lukaszsamson]

Tightened refinement to non_empty_list/non_empty_map for length/map_size(x) == n when n > 0. Added all the test cases suggested by @gldubc

[gldubc]

I think it's good now!

[josevalim]

💚 💙 💜 💛 ❤️