fix(dispatch-suite): watch runs at 60s to cut secondary-rate-limit load

[joshua-temple]

Problem

Concurrent fleet gates each ran gh run watch at its 3s default cadence. With up to 8 suites watched at once for multi-minute runs, that is hundreds of API calls per suite and tripped GitHub secondary rate limits, breaking runs.

Fix

• gh run watch now polls at a configurable interval via a new watch-interval input (default 60). For multi-minute suites, a 60s tick detects any state change within negligible delay while cutting watch API calls about 20x.
• recover-interval already defaults to 10s with 30 attempts (300s budget), so recovery headroom is unchanged.

Verification

• actionlint clean on the consumer workflow (fleet-e2e.yaml).
• Default applies automatically; no caller change required. gh run watch confirmed to support -i/--interval.