bug: nil UUID fallback for agentId causes silent event loss

[beonde]

Problem

In GuardEventEmitter.emit_policy_enforced(), when _resolve_agent_id() returns None, the code falls back to str(uuid.UUID(int=0)) (all-zeros UUID 00000000-0000-0000-0000-000000000000).

However, the server's IngestEvent handler explicitly rejects nil UUIDs:

if req.AgentID == uuid.Nil {
    response.BadRequest(w, "agentId is required")
    return
}

This means events are silently dropped whenever agent resolution fails. The logger.warning on the 400 response surfaces the symptom, but the telemetry data is permanently lost.

Where

capiscio_mcp/events.py — emit_policy_enforced() method, line:

event: Dict[str, Any] = {
    ...
    "agentId": effective_agent_id or str(uuid.UUID(int=0)),
    ...
}

Suggested fix

Either:

1. Skip emission entirely when no valid agentId can be resolved — log a warning and return early
2. Require a valid agent_id at GuardEventEmitter construction time (fail loudly at setup, not silently per-event)

Option 1 is simpler and maintains the best-effort contract. Option 2 is stricter but catches misconfiguration early.

Context

Introduced on feat/guard-event-emission branch as part of the event payload restructuring.