Instead of managing a large list of tests to skip as we break aarch64,
simply don't run any userspace tests. We need to keep a smoke test going
so that we don't break aarch64 boot with new Kernel changes and new
drivers.
This ensures that the RAM does not fill up with already processed
coredumps when many tests crash (as is the case on AArch64). We only
do this in self-test mode so as to avoid racing CrashDaemon.
This shouldn't be necessary but reproducing the odd corrupted profile
data from the nightly Azure runs is proving tricky locally. Hopefully
this will mitigate the issue.
Set LLVM_PROFILE_FILE to a pattern that is easily extractable as "these
are the profile data files we generated while running tests" for later
post-processing.