LLVM's smul.with.overflow intrinsic isn't supported on X86 for bit
widths larger than 64, or on X86-64 for bit widths larger than 128.
The failure mode is either a linker error ("the __muloti4 builtin isn't
available for this target") or an assertion failure ("SelectionDAG
doesn't know what builtin to call").
Until we actually add builtin support for 128-bit multiply-with-overflow
on X86, we should error-out on unsupported calls as early as possible.