fix: exempt non-autonomous agents from heartbeat inactivity timeout

[Fail-Safe]

Problem

Non-autonomous (reactive) agents were being flagged as unresponsive and crash-recovered after sitting idle for default_timeout_secs (180s), even though idle is their normal state. They wait for incoming messages and have no expected self-trigger schedule.

This caused unnecessary crash/recovery cycles for healthy agents that simply hadn't received a message in the last 3 minutes.

Root Cause

check_agents in heartbeat.rs applied the same inactivity timeout to all Running agents regardless of type. For agents without an autonomous config block, it fell back to config.default_timeout_secs (180s). A reactive agent idle for 3+ minutes would be indistinguishable from an autonomous agent that had stalled.

Fix

Make the inactivity check conditional on agent type:

• Autonomous agents retain the heartbeat_interval_secs × 2 inactivity check — meaningful because they are expected to fire periodically.
• Non-autonomous agents are only flagged when their state is Crashed; idle time is no longer checked at all.

// Before
let timeout_secs = entry_ref.manifest.autonomous.as_ref()
    .map(|a| a.heartbeat_interval_secs * UNRESPONSIVE_MULTIPLIER)
    .unwrap_or(config.default_timeout_secs) as i64;
let unresponsive = entry_ref.state == AgentState::Crashed || inactive_secs > timeout_secs;

// After
let timeout_secs: Option<i64> = entry_ref.manifest.autonomous.as_ref()
    .map(|a| (a.heartbeat_interval_secs * UNRESPONSIVE_MULTIPLIER) as i64);
let unresponsive = match timeout_secs {
    Some(t) => entry_ref.state == AgentState::Crashed || inactive_secs > t,
    None    => entry_ref.state == AgentState::Crashed,
};

Testing

Verified with a 5-agent setup (mix of autonomous and reactive). Before the fix, idle reactive agents were logged as unresponsive at inactive_secs=210. After the fix, they show heartbeat OK at 210s, 240s, and beyond with no crash/recovery cycles.

• cargo build --workspace --lib passes
• cargo clippy --workspace --all-targets -- -D warnings passes (zero warnings)
• cargo test --workspace passes
• Live integration tested

[jaberjaber23]

Real bug, clean fix, not slop. But needs tests before merge.

The fix correctly exempts reactive agents from inactivity timeouts — they're designed to sit idle waiting for messages. However:

1. No tests. The check_agents function is pure and trivially testable. Need at minimum: (a) reactive agent idle 5 min is NOT flagged unresponsive, (b) reactive agent in Crashed state IS flagged, (c) autonomous agent idle beyond timeout IS flagged.

2. default_timeout_secs becomes dead code after this PR — never read at runtime. Should document why it's retained or remove it.

3. The warn! macro now logs timeout_secs=Some(60) instead of timeout_secs=60. Should unwrap the Some in the log since it's guaranteed at that point.