Add error handling/retries to Go -> Docker auto-update infra

The go-docker auto-update infra has some potential for races: it first checks to see if a PR exists, and based on that, decides whether to fetch from an existing branch or create a new branch, and whether to ff push or force push to the branch. Even within a single decision, there is a short window for a race between the fetch and the push where other auto-updates could race.

We might be more likely to hit this than it seems: the branch sync runs on a schedule, so microsoft/main and microsoft/go.release-branch.* builds (and therefore Docker updates) will probably run around the same time, and eventually could happen to line up.

Mitigations:

• The Docker auto-update is quick (10s).
• I haven't thought of a condition where this problem will cause a bad change to get merged. I think we just might miss a particular update (which we can retry manually if we care about that update in particular).

We should adjust the auto-update code to account for this.

Start by locating the Go-to-Docker auto-update code and the path that checks for an existing PR before fetching, creating branches, or pushing. Review how scheduled updates can overlap; done means race-related fetch or push failures are handled with retries or clear errors without losing an update.