Drive open-task occupancy when the fleet dry edge is missed

[TheGreatAxios]

Summary

Occupancy takes one dry-episode shot when the parent settles idle, even if the live fleet 1→0 event was never observed. A no-op (no open tasks) does not consume the shot; a true drive still blocks the loop until a live lane resets.

Idle-with-fleet lets the parent go idle while workers run. CL-7540 latched the open-task continuation on live count going from N>0 to 0. If that edge is missed — parent already idle with a dry fleet, fleet events never showed a live count, or a continuation abort idles without consuming the latch — open tasks sit there and the session looks stalled.

• Idle + dry + todo/doing drives exactly once.
• A new live lane resets the latch so the next dry episode can fire once more.
• Idle + dry + open tasks does not start a continuation on every settle.
• Parent still processing, or a live lane, still blocks the drive.
• This is not CL-7518 (mailbox inbound). Wait timeouts are unchanged.

Verification

• bun run typecheck, bun run build, and bun run test pass
• bun run check passes
• bun test src/tui/runtime-bridge.test.ts --randomize --seed 424242 covers missed-edge drive-once, second settle no-op, live-lane reset, and parentProcessing refuse

Fixes CL-7625

[linear-code]

CL-7625

[TheGreatAxios]

Critic · Comment

Occupancy takes one dry-episode shot on idle settle even without a live 1→0; a live lane resets the latch; parentProcessing still refuses.

No findings.