don't show tracebacks unless all attempts are exhausted

[bhearsum]

In automation at Mozilla, we do a lot of heuristic log parsing to find exceptions and highlight them. One of the scenarios we've been running into a lot lately is where we have a job that uses redo, it prints some tracebacks, a subsequent attempt succeeds, but then the job fails for an unrelated reason (eg: a timeout is some non-retried thing). This ends up with the traceback that we successfully retried for getting highlighted, and confusion ensues.

We should stop printing tracebacks at less verbose log levels unless we've exhausted all the attempts.