Sentinel: Crew runs in trevato/quorum keep failing at the push or the sandbox #2035

This change only fixes how the failure is labelled. A crew run that ends on a network outage is still a failed run.

The runner already retries a dropped connection three times (5, 20 and 60 seconds). If the network is still down, it pauses the run for 5 minutes, up to MAX_PAUSES (6) times. After the last pause the run ended with no changes pushed: API Error … EAI_AGAIN. That is the line the sentinel counts as a push failure, and it points at the push when the real cause is the network.

What changed

• In src/crew/runner.ts, a run that ends on a network error with nothing landed now says network unreachable after 6 pauses; nothing to land (<reason>). Runs that did push, and non-network failures, keep their old summaries.
• In test/crew-network.test.ts, the exhausted-pauses test now checks for the new summary and that no changes pushed is absent.
• bun test test/crew-network.test.ts test/crew-parks.test.ts and tsc --noEmit pass. I did not run the whole suite; the forge's check does that on the push.

What I left out

• The DNS failure in the sandbox itself. I can't see or fix it from the repository, and the retry and pause logic already covers short outages.
• The earlier commits on this issue (the #2012 ones) carry no file changes, so there was no earlier fix to build on.

To try it: bun test test/crew-network.test.ts. The "outage that outlasts the retries" case covers the new summary.

Closes #2034