Reinstate retries for image pulls

[rnorth]

No description provided.

[rnorth]

@kiview if you could please review as well I'd be grateful. After @bsideup's review I'm afraid I made a couple of changes:

• I noticed that pull failures were still occuring in CI. This really shouldn't happen, but my hypothesis is that when CI->Docker Hub network glitches occur they are brief periods of 0% availability. 3 quick retries in these periods are pointless. So instead of this I've modified the retry logic to allow as many retries as necessary up to a maximum of 2 minutes. As a result, temporary network blips should get enough time to recover but there is a sanity limit for retries.

• I also noticed that the retries were a bit blunt - we were retrying in the case of authentication or 'not found' failures, which is clearly unhelpful. We obviously don't want to retry these for up to 2 minutes! Instead, I've changed the code to let these kind of failures fail-fast and bubble up the exception immediately - zero retries.

I hope this makes sense!

I've made significant changes after review

[bsideup]

👍