Recently, we had a lab outage caused when a large number
of Chrome crashes from testing in the lab overwhelmed
the results gathering/offloading system.  The crashes
didn't trigger test failures, probably they occurred
during browser restart, outside of testing.  This is
actually the second time we've seen such an event.

One of the observed problems in such events is that,
even when we know that the problem is Chrome crashes,
unless we can find test failures, we can't find the
test logs with the dumps.  This causes lots of problems:
  * Without the dumps, we can't get to the root cause.
  * Without the logs, we can't guess at how to reproduce
    a problem locally.
  * It may happen that the only reproducible case is in
    the lab, in which case, without logs, we can't do
    anything at all.

We need some way to be able to find crash dumps, independent
of whether there's a test failure that points to the dump.

One likely suggestion is to upload the crashes using the
standard product infrastructure.  See the discussion here:
    http://shortn/_r13oCGTQK6