To be clear, the dropbox violations were already async in the
ActivityManager, but the Binder call was often 30 ms anyway.
This optimization was already done for per-thread violations earlier,
but was never done for VM-wide violations because they weren't common,
until CloseGuard came about. Now that CloseGuard fires a lot, apply
the same optimization to VM-wide violations.
This CL also addresses a concern of Dianne's earlier of too many
threads being outstanding. So now there's a paranoia check with an
upper bound on how many outstanding ActivityManager calls are
in-flight.
Change-Id: I95e0816105ab862f0f241052b149c9a46a70ce9c
bee248769d