cross-crate calls to generic class methods yields unconstraind tvars

[nikomatsakis]

I recently made a change that reports errors for unconstrained type variables rather than resolving them to bottom (the latter was causing confusion). This seems to have unearthed a problem with cross-crate method calls on classes where we end up with unconstrained type variables. Specifically: the class type variables seem to be unresolved. I believe the problem has to do with the encoding, which encodes both the class tps and method tps together rather than just the method tps. But if I just revert this line, it causes failures in kind---my belief there is that there is some compensatory logic. But I didn't spend enough time to be sure, I could just be misunderstanding how it works. Anyway, for now I have xfailed src/test/run-pass/class-poly-methods-cross-crate.rs

[brson]

punting to 0.4

[nikomatsakis]

I had already fixed this.