perf(strm): 优化 OpenList 元数据下载速度 (#15)

* Add favourite buttons to library list and series detail pages

Co-authored-by: truewhile <truewhile@users.noreply.github.com>

* Fix favourites list for remote Emby mounted media

Co-authored-by: truewhile <truewhile@users.noreply.github.com>

* Route favourites to library series detail when appropriate

Co-authored-by: truewhile <truewhile@users.noreply.github.com>

* Fix favourites link to library series for remote anime paths

Co-authored-by: truewhile <truewhile@users.noreply.github.com>

* Fix remote Emby playback history and continue watching

- Preserve duration_ms when web player reports duration=0
- Add GET /playback/:id/resume for reliable per-item resume
- Switch home continue watching to /watch-history/continue API
- Set library_id on remote media for play-profile visibility
- Keep continue-watching rows when remote metadata hydration fails

Co-authored-by: truewhile <truewhile@users.noreply.github.com>

* Fix mobile sort dropdown overflowing off-screen

Align dropdown to button left edge on small screens and cap width
to viewport so sort options stay fully visible on mobile.

Co-authored-by: truewhile <truewhile@users.noreply.github.com>

* Remove stale Windows binary backup files

Co-authored-by: truewhile <truewhile@users.noreply.github.com>

* Fix STRM account edit form not echoing saved config

Return config_preview from account API for server/url/username and
secret flags. Merge partial config updates so empty password/token
fields do not wipe stored credentials.

Co-authored-by: truewhile <truewhile@users.noreply.github.com>

* perf(strm): speed up OpenList metadata downloads with provider-aware concurrency

- Split download semaphore: 115 stays capped at 3 for WAF safety; OpenList/CloudDrive2
  WebDAV metadata uses strm.download_threads (default raised to 6, max 16)
- Scope 115 WAF cooldown to 115 tasks only so OpenList downloads are not paused
- Requeue claimed tasks back to pending when slot acquisition fails or 115 is cooling
- Add tests for separate semaphores and requeue behavior

Co-authored-by: truewhile <truewhile@users.noreply.github.com>

---------

Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: truewhile <truewhile@users.noreply.github.com>

internal/service/strm_queue.go

@@ -29,9 +29,8 @@ const (
 
 // downloadWorker 下载队列 worker：认领 → 解析直链 → 下载 → 落盘。
 //
-// 采用「批量认领 + 全局并发限流」：一次认领数个任务，用 StrmService 上的全局信号量
-// 限制整个进程「同时换直链+下载」的并发数（与 115 换链风控匹配，见 strmDownloadSemCap），
-// 同时让下载充分并行。换链走全局令牌桶（QPS=3）兜底，下载走 CDN 不限速。
+// 采用「批量认领 + 按网盘类型限流」：一次认领数个任务，115 与 WebDAV/OpenList
+// 使用独立信号量（115 固定 3 防风控；OpenList/CloudDrive2 并发由 download_threads 控制）。
 func (s *StrmService) downloadWorker(ctx context.Context) {
 	const claimBatch = 12 // 每次批量认领的任务数
 	for {
@@ -42,12 +41,6 @@ func (s *StrmService) downloadWorker(ctx context.Context) {
 			return
 		default:
 		}
-		// 115 风控/限流熔断：冷却期间整体暂停，不给 WAF 续封机会
-		if left := s.wafCooldownLeft(); left > 0 {
-			s.log.Debug("下载队列冷却中", zap.Duration("remaining", left))
-			sleepContext(ctx, left)
-			continue
-		}
 		tasks, err := s.repo.StrmDownload.ClaimPendingDownload(ctx, claimBatch)
 		if err != nil {
 			s.log.Warn("claim strm download task failed", zap.Error(err))
@@ -58,24 +51,37 @@ func (s *StrmService) downloadWorker(ctx context.Context) {
 			sleepContext(ctx, 2*time.Second)
 			continue
 		}
-		// 并发处理本批任务：每个任务先获取全局下载槽位，槽位内部执行换链+下载。
-		// 信号量与令牌桶双重限速，确保任意时刻并发换链请求不超过安全阈值。
 		var wg sync.WaitGroup
 		for i := range tasks {
 			wg.Add(1)
 			go func(i int) {
 				defer wg.Done()
-				if !s.acquireDownloadSlot(ctx) {
+				task := &tasks[i]
+				if task.Provider == model.StrmProvider115 && s.wafCooldownLeft() > 0 {
+					s.requeueDownloadTask(task)
 					return
 				}
-				defer s.releaseDownloadSlot()
-				s.processDownloadTask(ctx, &tasks[i])
+				if !s.acquireDownloadSlot(ctx, task.Provider) {
+					s.requeueDownloadTask(task)
+					return
+				}
+				defer s.releaseDownloadSlot(task.Provider)
+				s.processDownloadTask(ctx, task)
 			}(i)
 		}
 		wg.Wait()
 	}
 }
 
+// requeueDownloadTask 把已认领但未实际执行的任务退回 pending，避免长期停留在 running。
+func (s *StrmService) requeueDownloadTask(task *model.StrmDownloadTask) {
+	task.Status = model.StrmTaskPending
+	task.StartedAt = nil
+	if err := s.repo.StrmDownload.Update(context.Background(), task); err != nil {
+		s.log.Warn("requeue strm download task failed", zap.Error(err), zap.String("id", task.ID))
+	}
+}
+
 func (s *StrmService) processDownloadTask(ctx context.Context, task *model.StrmDownloadTask) {
 	cleanPath := sanitizeLocalPath(task.LocalPath)
 	if cleanPath != "" && cleanPath != task.LocalPath {
@@ -648,7 +654,7 @@ func sleepContext(ctx context.Context, d time.Duration) {
 
 // ─── 115 风控/限流熔断 ────────────────────────────────────────────────────────
 
-// triggerWAFCooldown 检测到 115 风控/限流后触发全局冷却，冷却期间下载 worker 暂停。
+// triggerWAFCooldown 检测到 115 风控/限流后触发冷却，仅影响 115 下载任务。
 // 冷却时间取最大值，避免连续触发时缩短等待。
 func (s *StrmService) triggerWAFCooldown() {
 	s.mu.Lock()