Clean up resolve error reporting

[emberian]

Currently, resolve does a lot of redundant error reporting. For example,

/home/cmr/hacking/rust/src/libextra/interner.rs:161:8: 161:18 error: unresolved name
/home/cmr/hacking/rust/src/libextra/interner.rs:161         ast::ident { repr: off, ctxt: 0 }
                                                            ^~~~~~~~~~
/home/cmr/hacking/rust/src/libextra/interner.rs:161:8: 161:18 error: use of undeclared module `ast`
/home/cmr/hacking/rust/src/libextra/interner.rs:161         ast::ident { repr: off, ctxt: 0 }
                                                            ^~~~~~~~~~
/home/cmr/hacking/rust/src/libextra/interner.rs:161:8: 161:18 error: `ast::ident` does not name a structure
/home/cmr/hacking/rust/src/libextra/interner.rs:161         ast::ident { repr: off, ctxt: 0 }
                                                            ^~~~~~~~~~

This should be a single error, "error: use of undeclared module ast".

Another example:

/home/cmr/hacking/rust/src/libextra/interner.rs:161:8: 161:18 error: unresolved name
/home/cmr/hacking/rust/src/libextra/interner.rs:161         ast::ident { repr: off, ctxt: 0 }
                                                            ^~~~~~~~~~
/home/cmr/hacking/rust/src/libextra/interner.rs:161:8: 161:18 error: use of undeclared module `ast`
/home/cmr/hacking/rust/src/libextra/interner.rs:161         ast::ident { repr: off, ctxt: 0 }
                                                            ^~~~~~~~~~
/home/cmr/hacking/rust/src/libextra/interner.rs:161:8: 161:18 error: `ast::ident` does not name a structure
/home/cmr/hacking/rust/src/libextra/interner.rs:161         ast::ident { repr: off, ctxt: 0 }
                                                            ^~~~~~~~~~

Fixing this will require careful finagling of resolve. The best way to handle this is probably some central error reporting infrastructure, and as soon as one error is reported, stop making new errors for the current span. This is going to be tricky because every error path needs to be followed to ensure that an error is reported, and that if only one error is reported, it is the right error. Needs vastly improved test coverage in compile-fail. Shouldn't be hard, but will take time.

Nominating for production-ready.

[catamorphism]

I actually implemented a change similar to this in my fix for #6702. I think there are probably some duplicate errors still, but the infrastructure is now there for doing what this bug suggests (I added a flag in the context saying whether or not to emit errors, and a method on the context to run a closure without emitting any errors).

[catamorphism]

I think my changes provide support for doing this -- feel free to open another bug with specific test cases where it still goes wrong. Closing.