Unreliable buildbots

There's a thread discussing LLDB buildbot unreliability taking place
on llvm-commits. I just noticed that lldb-dev isn't on the CC. The
discussion started on the topic of the FreeBSD LLDB builder but
progressed to discussing the general unreliability across the bots.


Should all of the LLDB bots (perhaps other than the Windows build-only
bots) be moved to the experimental category so that they don't
generate failure emails? Should we have (in addition) build-only
configurations for Darwin/FreeBSD/Linux?