Add atomic task deduplication with partial unique index

- Add partial unique index for active resource-scoped tasks
- Implement createTaskRunRecordAtomic for race-free task creation
- Update findOrEnqueueTask to use atomic insert first
- Add tests for concurrent task creation deduplication

lib/server/db/schema.ts

@@ -668,6 +668,12 @@ export const taskRun = sqliteTable('task_run', {
   taskWorkflowRunUnique: uniqueIndex('task_workflow_run_uidx').on(table.workflow_run_id)
 }));
 
+// Note: Partial unique index for active resource-scoped task deduplication is created via
+// migration 0013_task_active_resource_unique.sql. SQLite does not support partial indexes
+// in drizzle schema DSL, so the index is managed separately:
+// CREATE UNIQUE INDEX task_active_resource_uidx ON task_run (user_id, task_type, resource_key)
+// WHERE resource_key IS NOT NULL AND status IN ('queued', 'running');
+
 export const taskStageEvent = sqliteTable('task_stage_event', {
   id: integer('id').primaryKey({ autoIncrement: true }),
   task_id: text('task_id').notNull().references(() => taskRun.id, { onDelete: 'cascade' }),