fix(agent-teams): keep in-flight turns from failing when the store closes during shutdown

[Ivancheng7]

Problem

AgentTeamStore.close() closes the SQLite connection during shutdown, but a turn that was already running may still be finalizing its card. The final-card path calls enrichWithAgentTeams(), which reads findTeamForChat() and status(). Once the connection is closed those reads throw The database connection is not open, so the turn is reported as task_error — exactly the turns that were in flight when the process restarted, i.e. the ones a restart should let finish.

Seen on a systemd-managed deployment: a systemctl restart window is enough to fail whatever turn was running at that moment.

Fix

• AgentTeamStore tracks a closed flag: listTeams() / getTeam() (and everything built on them) return "no data" after close instead of throwing; close() is idempotent; isClosed() is exposed for callers.
• enrichWithAgentTeams() skips enrichment once the store is closed, and treats any remaining read failure as auxiliary — the card renders without the team block instead of failing the turn.

The narrow race between the isClosed() check and the read remains (close() can land in between), which is why the caller treats a read failure as non-fatal rather than relying on the flag alone.

Tests

tests/agent-team-store.test.ts adds: read accessors after close, idempotent close. The full local suite on this branch matches upstream/main — no new failures.