feat: 加固运行期可靠性并补齐调度与观测接口

- 运行句柄按 run_id 归属，修复取消后重跑时闸门被旧运行删除的竞态
- 调度循环改用取消令牌退出，启动时对账遗留运行，校验 rows_affected
- 三类任务统一参与调度，新增 /api/v1/stats 与 /api/v1/schedule
- SQLite 启用 WAL 与 busy_timeout，日志清理跳过活跃文件，单行上限 256 KiB
- 前端缓存管理令牌并在 403 时重试一次，仪表盘改用统计接口

backend/src/runs/repository.rs

@@ -119,20 +119,22 @@ pub async fn create_skipped(
     })
 }
 
+/// 返回是否真的更新到了运行记录；记录被删除（例如任务被删）时返回 false。
 pub async fn mark_running(
     pool: &SqlitePool,
     run_id: i64,
     started_at: DateTime<Utc>,
-) -> ApiResult<()> {
-    sqlx::query("UPDATE task_runs SET status = ?, started_at = ? WHERE id = ?")
+) -> ApiResult<bool> {
+    let result = sqlx::query("UPDATE task_runs SET status = ?, started_at = ? WHERE id = ?")
         .bind(RunStatus::Running.as_str())
         .bind(started_at)
         .bind(run_id)
         .execute(pool)
         .await?;
-    Ok(())
+    Ok(result.rows_affected() > 0)
 }
 
+/// 返回是否真的更新到了运行记录。
 pub async fn complete(
     pool: &SqlitePool,
     run_id: i64,
@@ -141,8 +143,8 @@ pub async fn complete(
     duration_ms: u64,
     exit_code: Option<i32>,
     error_message: Option<&str>,
-) -> ApiResult<()> {
-    sqlx::query(
+) -> ApiResult<bool> {
+    let result = sqlx::query(
         "UPDATE task_runs SET
             status = ?, finished_at = ?, duration_ms = ?,
             exit_code = ?, error_message = ?
@@ -156,7 +158,22 @@ pub async fn complete(
     .bind(run_id)
     .execute(pool)
     .await?;
-    Ok(())
+    Ok(result.rows_affected() > 0)
+}
+
+/// 启动时收敛上一次进程遗留下来的非终态运行，避免界面永远显示"运行中"。
+pub async fn fail_interrupted_runs(pool: &SqlitePool) -> ApiResult<u64> {
+    let result = sqlx::query(
+        "UPDATE task_runs SET status = ?, finished_at = ?, error_message = ?
+         WHERE status IN ('queued', 'running')",
+    )
+    .bind(RunStatus::Cancelled.as_str())
+    .bind(Utc::now())
+    .bind("TaskPilot 进程重启，运行被中断")
+    .execute(pool)
+    .await?;
+
+    Ok(result.rows_affected())
 }
 
 pub async fn get(pool: &SqlitePool, run_id: i64) -> ApiResult<Option<TaskRunResponse>> {