GitLab

Crash was cauded by a null experiment, which is something we handled
elsewhere but not in this particular code. Also adds a unit test is
able to trigger the crash before the fix, although I'm not very
convinced this is the exact edge case from the wild (but either
way the fix should apply to both).

Also adds some comments about some code that was a bit confusing.
One is about using the previous group for kill switch type and
another about the threading when simulation.

Bug: 837540
Change-Id: If93f4658cd18019f3f40c82ec3172d10f61356f3
Reviewed-on: https://chromium-review.googlesource.com/1036104Reviewed-by: Ilya Sherman <isherman@chromium.org>
Commit-Queue: Alexei Svitkine <asvitkine@chromium.org>
Cr-Commit-Position: refs/heads/master@{#556115}