fix: reset last_active on agent restore to prevent heartbeat false-positives on startup

[Fail-Safe]

Problem

Every daemon restart triggers a cascade of spurious crash-recovery cycles for all agents.

When agents are loaded from persistent storage, their last_active timestamp is restored from before the previous shutdown. If the daemon was down for longer than the heartbeat unresponsiveness threshold (default 180 s), the very first heartbeat tick — which fires ~30 s after startup — sees every restored agent as stale and immediately marks them all as Crashed, kicking off recovery attempts for agents that had only just been restored and hadn't had a chance to run.

Observed in logs:

INFO  openfang_kernel: Restored 6 agent(s) from persistent storage
...30 seconds later...
WARN  openfang_kernel::heartbeat: Agent is unresponsive agent=Athena inactive_secs=210 timeout_secs=180
WARN  openfang_kernel::heartbeat: Agent is unresponsive agent=Paige  inactive_secs=210 timeout_secs=180
WARN  openfang_kernel::heartbeat: Agent is unresponsive agent=Cortex inactive_secs=210 timeout_secs=180
WARN  openfang_kernel::kernel:    Unresponsive Running agent marked as Crashed for recovery agent=Athena
...

The recovery succeeds each time, but the unnecessary crash-recovery round-trip adds noise to logs, delays agent availability by 30–60 s on every restart, and burns recovery attempt budget.

Fix

One line added to the agent restore loop in kernel.rs, alongside the existing state = AgentState::Running reset:

restored_entry.last_active = chrono::Utc::now();

This gives each restored agent a fresh heartbeat baseline consistent with how newly spawned agents are initialized (they also set last_active = Utc::now()). The heartbeat monitor then starts timing from the actual moment of restoration rather than from the pre-shutdown timestamp.

Impact

• No behavioural change for agents that are genuinely unresponsive after restoration
• Agents that fail to start their loop after restore will still be caught by the heartbeat on the next check cycle (30 s later)
• Eliminates the false-positive crash storm on every daemon restart