Restore complete planned-restart chat recovery

[hamzamerzic]

Summary

• remove the process-wide two-chat admission gate from authenticated planned-restart continuation
• restore every eligible pre-restart chat in the same bounded continuation sweep
• retain nonce, policy, latest-run, unanswered-question, and queued-app-work safeguards

Why

#516 introduced a global live-chat ceiling to keep restart recovery from competing with new shell work. The ceiling only admitted another parked chat after the total live count fell below two. Recovered agent turns can remain active for a long time, so later chats could stay parked indefinitely even though Continue after planned restarts was enabled.

The planned-restart contract established by #396 restores the exact authenticated set that was already concurrent before the restart. This focused change reinstates that contract. Provider-limit continuation remains separately staggered and starts at most one chat per sweep.

Safety

Automatic continuation still requires the current boot's supervisor-authenticated nonce, the latest exact restart park, the chat-local restart policy, and no unanswered question. New app-attributed work is still excluded. Deleted, superseded, or manually transitioned chats remain on their existing paths. The existing oldest-first sweep batch remains bounded.

Testing

• restart and continuation suites: 92 passed
• hermetic backend contract gate: 69 passed
• Python compilation and diff checks passed
• the complete Docker backend gate is unavailable inside the live app container; hosted CI remains authoritative