fix re-try to surpress timeout warnings when they're dealt with

[zanerock]

If you do the following, about 1 in 100 will fail with what looks like a timeout problem:

COUNT=0; while [ $COUNT -lt 300 ]; do echo $COUNT; if ! bash tests/unit/topics/submit/current-topic.sh; then COUNT=1000; fi; COUNT=$(($COUNT + 1)); done

28
ERROR: 'con topics submit' was not expected to produce error output, but got:
ERROR: REST call failed for 'POST /repos/DogFoodSoftware/test-repo/pulls {"title": "Pull request for issue #1161","body": "Fixes #1161. Generated by conveyor-workflow.","head": "topics-1161-b565b9ab-35a8-49d9-a348-9bd4559cde52","base": "master"}'. (200)
29

I think the first 28 call times out. The '(200)' is because the previous call did succeed. We're either not re-trying timed out calls or we are, but leaking the error prematurely. Could be something else, but that's where to start.

[zanerock]

Yes, if you look at the code in github-hooks.sh#create_pull_request(), you can see it tests on explicit error messages, the HTTP status, but not the return status. This was because we didn't want to retry for other HTTP problems, but if there is no HTTP status, and the call failed, we can take that as a timeout and try again. (After we fix the problem with not clearing the previous HTTP status.)