fix(fleet): tolerate transient watch API errors with a bounded retry

[joshua-temple]

Problem

The fleet dispatch-suite action and the auto-promote Release watcher used gh run watch --exit-status, which fails the whole job on a single transient HTTP 401/403/5xx mid-poll even when the watched run concludes success. This reded a fully-green fleet run (rc.4 primary) on a transient 401 under load.

Fix

Both watch sites now use a bounded gh run view --json status,conclusion poll loop that:

• retries transient API errors (401/403/5xx/rate-limit/network), with a consecutive-error cap that fails closed after repeated failures;
• exits 0 only on status=completed AND conclusion=success;
• exits 1 on any other concluded conclusion (never masks a real failure);
• fails closed on a stuck run via a ~30-minute wall-clock cap.

.github-only change (no rc cut). actionlint clean. Closes #335.