Fix some dead-code pass bugs

[ktt3ja]

Fix #10865 and #10939.

[alexcrichton]

Special casing on an arm is a little too special-casing for me. This has nothing to do with arms of a match, but rather something which can show up in patterns.

I believe the correct solution for this is in #10862 where a visit_path method was added. The real bug here is that paths in patterns aren't being matched, not that patterns in arms aren't being matched. For now, let's wait for the outcome of #10862 and then rebase this on top of that (and override visit_path in the visitor).

I'd also be fine if you wanted to rebase that patch in this PR and I'll r+ this one as soon as that one gets r+'d (to get the rebase conflicts out of the way sooner).

[alexcrichton]

Now that my PR has landed, this should be ready for a rebase.

[ktt3ja]

I'll get on it on Wednesday after my exam.

[ktt3ja]

@alexcrichton: r?