store: Retry registerQueue on failure; start testing UpdateMachine.load

[gnprice]

Fixes #556.

In addition to the test this adds, I wanted to manually test it
end-to-end, to help be sure this covered the scenario where this
retry is known to be needed in the wild:

• The app is offline for a while, perhaps because the device
  is asleep.
• The app comes online, tries polling, and finds the event queue
  has expired, so it attempts a re-register.
• Before that completes (which after all takes several seconds
  if the realm is a large one), the app goes offline again.
• That request's response therefore never reaches the app,
  and so when it eventually comes back online it needs to retry.

Step 1 is annoying to carry out literally, because it means
waiting 10 minutes for the event queue to expire. To work around
that, I sabotaged the getEvents binding function to use a wrong
queue_id value:

'queue_id': RawParameter('wrong' + queueId),

so that the server would always respond with a BAD_EVENT_QUEUE_ID
error, just the same as if the queue had expired.

Then to take the app offline and online again, I just turned
airplane mode on and off on my device. Because I used a
physical device connected to my computer over USB, that caused
no interference to my watching the logs on the console.

In my manual testing, the retries worked perfectly: no matter
how many times I turned airplane mode on and off, or with what
timing, the app always returned to getting a fresh queue and
polling it for events.

[chrisbobbe]

Thanks! LGTM; please merge at will after seeing a small comment on a comment.

[chrisbobbe]

I think I don't yet understand the "doesn't fail the test" part. If the polling retry were removed, would the test fail, and if so, why?

I notice that the test does fail if I uncomment this line, and I think that makes sense; it's explained in your comment by "it clobbers […]".

[gnprice]

Here's a revised version; does this successfully explain it?

      // TODO UpdateMachine.debugPauseLoop is too late to prevent first poll attempt;
      //    the polling retry catches the resulting NetworkException from lack of
      //    `connection.prepare`, so that doesn't fail the test, but it does
      //    clobber the recorded registerQueue request so we can't check it.
      // checkLastRequest();

Specifically if you bypass the retry like so:

           case Server5xxException() || NetworkException():
+            rethrow;

then this test does fail. (As do the two tests specifically for that retry.)

[gnprice]

Thanks for the review! Merged, with that change to the comment.