fix(ui): failureReason "usage_limit" nachziehen + Worktree-Gate prüft die Platte

Zwei liegengebliebene Consumer aus den letzten beiden Commits:

- TaskRunner klassifiziert seit 07dd7570 "usage_limit", aber weder
  TaskRowViewModel.FailureReasonLabel noch vm.failureReason (de/en) noch
  die get_task-Tool-Beschreibung kannten den Wert — die UI zeigte
  "Grund unbekannt", das MCP-Doc listete weiterhin max_turns|timeout|error.
- TaskRowViewModel.CanOpenWorktree prüfte nur auf einen nicht-leeren
  String. Die Zeile behält den Path eines gemergten/verworfenen Worktrees,
  also war der Menüpunkt aktiv und Process.Start warf in den Footer.
  Jetzt zusätzlich Directory.Exists — dieselbe Prüfung, die
  WorktreesOverviewModalViewModel und MergeSectionViewModel schon machen.

src/ClaudeDo.Ui/ViewModels/Islands/TaskRowViewModel.cs

@@ -180,7 +180,11 @@ public sealed partial class TaskRowViewModel : ViewModelBase
     // placeholder so the menu wiring is uniform; always null until a real gate exists.
     public string? QuickSessionDisabledReason => null;
 
-    public bool CanOpenWorktree => !string.IsNullOrWhiteSpace(WorktreePath);
+    // The row keeps a merged/discarded worktree's recorded Path, so a non-empty string is not
+    // proof the folder is still there — same on-disk check the worktrees overview and the merge
+    // section do before offering "open". Only read when the context menu is built, not per row.
+    public bool CanOpenWorktree =>
+        !string.IsNullOrWhiteSpace(WorktreePath) && System.IO.Directory.Exists(WorktreePath);
     public string? OpenWorktreeDisabledReason => CanOpenWorktree ? null : Loc.T("tasks.reasonNoWorktree");
 
     public string? RefineDisabledReason
@@ -233,11 +237,12 @@ public sealed partial class TaskRowViewModel : ViewModelBase
     public bool HasFailureReason => Status == TaskStatus.Failed;
     public string FailureReasonLabel => FailureReason switch
     {
-        "max_turns" => Loc.T("vm.failureReason.maxTurns"),
-        "timeout"   => Loc.T("vm.failureReason.timeout"),
-        "cancelled" => Loc.T("vm.failureReason.cancelled"),
-        "error"     => Loc.T("vm.failureReason.error"),
-        _           => Loc.T("vm.failureReason.unknown"),
+        "max_turns"   => Loc.T("vm.failureReason.maxTurns"),
+        "timeout"     => Loc.T("vm.failureReason.timeout"),
+        "usage_limit" => Loc.T("vm.failureReason.usageLimit"),
+        "cancelled"   => Loc.T("vm.failureReason.cancelled"),
+        "error"       => Loc.T("vm.failureReason.error"),
+        _             => Loc.T("vm.failureReason.unknown"),
     };
     // max_turns gets the actionable detail (turns used/configured) since that's the one case
     // where the fix is "raise maxTurns and continue_task", not "reset and re-run".