[ClangLinkerWrapper] Emit extra job to generate -save-temps assembly (#224446)
Summary:
This is an oft-requested feature and unfortunately cannot be done
easily. The approach is just to make a redundant job that does the same
handling, but emits an `--lto-emit-asm` job in the background. Ideally
these get enough parallelism that the total compile time doesn't change
much.
I cannot find a cleaner way to return the 'old driver behavior' for
emitting assembly as was requested. THe LTO pipeline is not built to
generate intermediate files like this, and running this *then* consuming
the assembly on output would break thinLTO, partitioned LTO, or future
object linking.