test/run-pass/issue-2316.rs fails if ran without a full make first

The tests no longer depend on librustc. This one shouldn't either. It is easy to reproduce the situation that led to the bug in a module inside the test program, there's no need to reach into the internals of librustc (in fact, that's very fragile, and might even result in the test passing when the bug resurfaces, since the imports in rustc/middle/ty.rs can change at any time).
