fix(team): converge run-scoped wakes into a run at the enqueue choke-point

[piorpua]

Summary

Fixes Sentry ELECTRON-3RC ("团队协作: no active team run", reported on AionUi 2.1.41).

Agent-initiated, run-scoped wakes (team_send_message and the same-source team_shutdown_agent) issued during a run-less window no longer hard-reject with no active team run for run-scoped wake (mapped to RuntimeContextMissing) or silently enqueue run-less. They now converge into an active SystemLifecycle team run at the single enqueue choke-point.

This closes the residual variant left open by #680 (38facd70), which converged system/lifecycle wakes but kept the guard as a probe and did not cover the agent-initiated path.

Root cause

A user message opens a team run and the leader enters a long turn. When every batch of that run is judged terminal, apply_work_summary's no_work branch marks the run Completed, so current_active_run_id() returns None while the leader agent is still inside its turn. In this run-less window:

• team_send_message (→ send_agent_message_from_agent) was hard-rejected by the service-layer guard require_active_team_run_for_team_work.
• Even without the guard, the agent→agent binding CausalBinding::InheritRunningBatch resolved to team_run_id: None in coordinator.rs::acquire_enqueue when there was no inheritable active batch, so no new run was opened.
• The same-source team_shutdown_agent used the same binding but skipped the guard, so it silently ran run-less.

A correct fix therefore had to address both the guard rejection and the binding that failed to open a run.

What changed (5 files, +138 / -35)

Production code:

• crates/aionui-team/src/work_coordinator/coordinator.rs — in acquire_enqueue, the CausalBinding::InheritRunningBatch branch now falls back to bind_system_enqueue when there is no inheritable active run (attach-or-create semantics, matching the adjacent SystemInitiated branch). Behavior is unchanged when there is an inheritable batch; only this one branch changed.
• crates/aionui-team/src/service.rs — removed the now-obsolete require_active_team_run_for_team_work guard: both its call in send_agent_message_from_agent and the function definition. With the choke-point guaranteeing a run on enqueue, the guard no longer has meaning.

Tests:

• crates/aionui-team/src/session.rs — updates confined to mod tests (no production change); two existing unit tests updated to the new invariant and renamed.
• crates/aionui-team/src/team_run/tests.rs — new unit test plus an updated assertion in mcp_message_inherits_the_callers_running_batch_causality.
• crates/aionui-team/tests/session_service_integration.rs — the former "requires active team run" send test converted to a GREEN regression (renamed); new same-source shutdown regression test added.

Testing

Run in the AionCore worktree; all green:

• cargo test -p aionui-team — 410 lib tests + all integration suites pass, 0 failed.
• cargo clippy -p aionui-team -- -D warnings — clean.
• cargo fmt --all -- --check — no diff.
• Full pre-push gate (just push) passed: 7480 tests run, 7480 passed, 18 skipped.

Scope and risk

• No schema/migration impact: team run state is in-memory only.
• Scope is intentionally bounded. The underlying run/turn lifetime mismatch (apply_work_summary marking a run Completed while the agent is still in a turn) and the separate team_task_update "Task not found" issue are explicitly out of scope; the latter is tracked separately.
• Residual risk is narrow: a future tool that mis-tags run-scoped work as Background would not be covered by bind_system_enqueue. This is a single-enum-choice review point rather than scattered guards.
• Manual verification anchor (not covered by automated tests): a real team session where the leader wakes a teammate after the user run completes.