Escalate publish pipeline failures

[cavis]

We've hit cases where failed publish attempts aren't loud enough. Or are too loud (we expect some failures/retries, uploading to Apple).

We should escalate errors based on how many / how long we've been failing.

• [ ] For <= 5 or something, just ops-warn
• [ ] For > 5, ops-error
• [ ] For > some number (hours?) ... ops-fatal