feat: F019 AI monitoring layer — output transparency & governance workspace

Full 2-panel workspace at /ops/ai-monitoring: metrics strip (total, failed,
pending, approval rate, top prompt version, active model), filterable output
table by type/status/error, and detail panel with output preview, prompt
versioning, error details, and role-aware review actions.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

src/components/ai-monitoring/AIErrorBadge.tsx

@@ -0,0 +1,23 @@
+import { Chip, Tooltip } from '@mui/material'
+import type { AIOutputError } from '../../domain/aiOutput'
+
+const ERROR_CONFIG: Record<string, { label: string; color: string }> = {
+  SCHEMA_VALIDATION: { label: 'Schema',     color: '#ea580c' },
+  PROVIDER_TIMEOUT:  { label: 'Timeout',    color: '#c0392b' },
+  INVALID_JSON:      { label: 'JSON',       color: '#c0392b' },
+  EMPTY_RESPONSE:    { label: 'Leer',       color: '#d97706' },
+  RATE_LIMIT:        { label: 'Rate Limit', color: '#7c3aed' },
+}
+
+export function AIErrorBadge({ error }: { error: AIOutputError }) {
+  const { label, color } = ERROR_CONFIG[error.type] ?? { label: error.type, color: '#c0392b' }
+  return (
+    <Tooltip title={error.message} arrow>
+      <Chip
+        size="small"
+        label={label}
+        sx={{ bgcolor: `${color}18`, color, fontWeight: 700, fontSize: '0.65rem', cursor: 'default' }}
+      />
+    </Tooltip>
+  )
+}