Repair Gmail arXiv replay reliability

[witbrock]

What changed

• de-route the obsolete broad mail-review workflow and make the Gmail/arXiv path converge through canonical paper-state inspection
• skip messages with verified processing markers, reuse represented papers, and represent only genuinely missing papers
• record and read back source-processing markers without mutating Gmail labels
• keep Google API connection fallback local to Gmail, preferring reachable IPv4 while retaining IPv6 fallback
• prevent the database-info probe from changing the process-wide socket timeout
• expose typed Gmail OAuth, scope, and transport failures to Von
• add focused workflow, Gmail transport, diagnostic, and error-classification coverage

Why

The replay was being derailed by a legacy workflow and later stalled before Gmail could return a response because the HTTP transport exhausted its timeout on unusable IPv6 addresses. The generic Gmail error envelope also made OAuth and transport failures too easy to conflate.

Validation

• live MJW Personal Chrome replay: durable instance 76e46041-5949-446f-8c6d-0d5a2ae08237 completed 25/25 message iterations with zero errors
• newest 20 messages skipped via existing markers; next five processed backwards and canonically read back
• no Gmail label mutation capability invoked
• 44 workflow/Gmail/settings focused tests passed
• 23 internal Gmail catalogue tests passed
• targeted Ruff checks and git diff --check passed

Atlas/start-up latency is deliberately out of scope.