usePerformanceAlerts now fetches /api/supervisor/performance-alerts
every 60s instead of computing client-side. Dismiss + dismiss-all hit
the sidecar so state survives reload. Toast fires when new alerts arrive.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
1. notification-bell: drop the DEMO_ALERTS fallback (Riya Mehta etc.).
Empty state ("No active alerts") shows when the live computation
returns nothing — which is the truthful state until thresholds are
set on Agent records.
2. use-performance-alerts: bucket calls by c.agentId === agent.id when
the relation is set; fall back to legacy agentName matching only for
un-enriched rows. Fixes conversion% calc going to 0 after backfill.
3. agent-table: Link target uses agent.id (UUID or "legacy:NAME") so
the URL is a stable identifier instead of a display string.
4. agent-detail: parse the route param into UUID vs legacy:NAME, filter
calls by c.agentId or c.agentName accordingly, and resolve display
name via the platform Agents list.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>