fix(signal): pin idle-liveness turn-lifecycle parity (no wall-clock cap)

[rysweet]

What

Locks the Signal channel to the idle-liveness turn lifecycle (no wall-clock per-turn timeout), completing the cross-transport parity tracked by #2604 / #2607 and finishing the work PR #2608 started for the meeting surface.

Why

The idle-liveness / no-wall-clock-timeout turn-lifecycle fix had to span dashboard, meeting, AND signal (#2604). Meeting got the hours-scale idle default in #2608; dashboard and signal route through the same MeetingBackend → PersistentAgentProxy path.

On main, Signal already inherits idle-liveness: open_signal_agent_session() builds its SessionBuilder in OperatingMode::Meeting, and Meeting mode is the seam that routes turns through PersistentAgentProxy — which reaps the agent child only on idle-liveness (no output for a generous window; every streamed chunk resets the clock) and never on a wall-clock elapsed-time cap. SIMARD_MEETING_IDLE_LIVENESS_SECS (default 3600s, per #2608) governs it, 0 = unbounded.

The gap #2604/#2607 leaves open is that nothing locked that wiring. A future refactor of the signal agent session onto a non-Meeting mode would silently drop it back onto the per-turn adapter and reintroduce the wall-clock turn timeout #2607 forbids — with no test to catch the regression.

Scope

• src/signal_conversation/channel.rs — extract the mode decision into a documented signal_agent_mode() returning OperatingMode::Meeting, with rationale citing Chat-management turn-lifecycle fix (idle-liveness + streaming) must span dashboard, meeting, AND signal — capture in proper worktrees/PRs #2604/No wall-clock timeouts on agent turns/subprocesses — reap on idle-liveness only (audit + signal parity) #2607; call it from open_signal_agent_session.
• src/signal_conversation/tests.rs — hermetic regression test signal_opens_agent_in_meeting_mode_for_idle_liveness_parity pinning the mode. If someone flips signal off Meeting mode, this fails loudly and points back to the parity requirement.

No behavioral change: signal turn handling already runs through the idle-liveness proxy today. This makes the guarantee explicit and regression-proof — the missing "Signal parity + test" acceptance criterion of #2604/#2607.

Tests

• cargo test --release --lib signal_conversation:: — 89 passed (incl. the new test)
• cargo clippy --release --no-deps -- -D warnings — clean
• cargo clippy --all-targets --all-features --locked -- -D warnings (pre-push) — clean
• cargo fmt --all -- --check — clean

Refs #2607, #2604. Follow-up to #2608, #2581, #2612.

Co-authored-by: Copilot 223556219+Copilot@users.noreply.github.com

[github-actions]

📊 Coverage Summary

Generated by cargo llvm-cov --workspace --summary-only (nightly, excluding test files)

Module	Lines	Covered	Coverage
Total	136694	113170	82.8%

_{Coverage data from CI run. Test files matching tests?/ are excluded from line counts.}