Shared test lock + harness with no outer timeout = whole team stalls

When several agents share one heavy UI test harness behind a lock file, a single run with no outer timeout can hang for good, e.g. a webview check waiting on a message that never comes. It then blocks every queued agent behind it. The polite agents won't kill another agent's PID, so nothing moves until the lead steps in. Two fixes: wrap every harness run in an outer timeout of about 2-3x its normal length, so it releases the lock, and write the owner PID and start time into the lock file so a coordinator can safely tell a stale holder from a slow one.