Unify random timeouts throughout LLDB and make them configurable.
Since these timeouts guard against catastrophic error in debugserver,
I also increased all of them to the maximum value among them.
The motivation for this test was the observation that an asanified
LLDB would often exhibit seemingly random test failures that could be
traced back to debugserver packets getting out of sync. With this path
applied I can no longer reproduce the one particular failure mode that
I was investigating.
Differential Revision: https://reviews.llvm.org/D60340