GitLab

NEVER_FINISHED is one of the larger (and growing :'( ) error buckets.
Add timeouts to every step to ensure that hangs are caught, turned into
errors for the right state, and the user has a chance to retry.

To get there we:
1. Add some timeouts, initially set to quite generous values which we'll
   tighten in future CLs. We have crosbolt data, but would like some
   real-world timings before getting stricter.
2. Add a metric tracking the time  spent in each step, to give us a
   real-world breakdown to supplement what we have in crosbolt
3. Add a metric tracking how often we see invalid transitions. In theory
   a slow operation could send its response during the next restart
   attempt, and since there's no validation on the state transitions
   we'd end up in a weird state with unknown effects. There are options
   with varying thoroughness (and corresponding cost) to address this,
   but first let's see if it's actually a problem. This could already
   happen (though we haven't had reports) and there are some mitigations
   in place already so it may be a non-issue.

Bug: 1142319
Test: Unit tests, mannually run installation + app launch.
Change-Id: Iafd15cf8f659b9a29764ce399359a29711e35043
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2563042
Commit-Queue: David Munro <davidmunro@google.com>
Reviewed-by: Mark Pearson <mpearson@chromium.org>
Reviewed-by: Ryo Hashimoto <hashimoto@chromium.org>
Reviewed-by: Dominick Ng <dominickn@chromium.org>
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Reviewed-by: Nicholas Verne <nverne@chromium.org>
Reviewed-by: Fergus Dall <sidereal@google.com>
Cr-Commit-Position: refs/heads/master@{#836505}