We do +assert builds of clang. Historically, this meant that clang emits somewhat verbose llvm code as well, but clang r263394 added a cc1 option to toggle this independently of +asserts.

Since more detailed llvm ir doesn't really have much to do with asserts, and since not emitting more detailed llvm ir is probably faster, we might want to measure if passing this flag helps any, and if so, pass it.