When we see something like i<i, we try to check whether the second "i" refers to a type. However, when we do this, we can end up creating an expression in the wrong context: the "<" has different parse priority if it's actually a relational operator.
To fix the issue, just remove the check. This makes diagnostics slightly worse, but avoids the crash. I'm not really happy with this, but I don't see another way to fix this without implementing a giant refactoring.
Fixes https://bugs.llvm.org/show_bug.cgi?id=43080 (but not the issues currently marked as "duplicate"; I haven't dug deeply into the causes of those issues).