feat: systematic functional audit — fix 18 issues across Phase A/B

Phase A (P1 production blockers):
- A1: Apply IP rate limiting to public routes (login/refresh)
- A2: Publish domain events for workflow instance state transitions
  (completed/suspended/resumed/terminated) via outbox pattern
- A3: Replace hardcoded nil UUID default tenant with dynamic DB lookup
- A4: Add GET /api/v1/audit-logs query endpoint with pagination
- A5: Enhance CORS wildcard warning for production environments

Phase B (P2 functional gaps):
- B1: Remove dead erp-common crate (zero references in codebase)
- B2: Refactor 5 settings pages to use typed API modules instead of
  direct client calls; create api/themes.ts; delete dead errors.ts
- B3: Add resume/suspend buttons to InstanceMonitor page
- B4: Remove unused EventHandler trait from erp-core
- B5: Handle task.completed events in message module (send notifications)
- B6: Wire TimeoutChecker as 60s background task
- B7: Auto-skip ServiceTask nodes instead of crashing the process
- B8: Remove empty register_routes() from ErpModule trait and modules

apps/web/src/api/errors.ts

@@ -1,13 +0,0 @@
-/**
- * Extract a user-friendly error message from an Axios error response.
- *
- * The backend returns `{ success: false, message: "..." }` on errors.
- * This helper centralizes the extraction logic to avoid repeating the
- * same type assertion in every catch block.
- */
-export function extractErrorMessage(err: unknown, fallback = '操作失败'): string {
-  return (
-    (err as { response?: { data?: { message?: string } } })?.response?.data
-      ?.message || fallback
-  );
-}