Add recovery for orphaned jobs

internal/server/ws/handler.go

@@ -357,10 +357,26 @@ func dispatchAgentMessage(ctx context.Context, c *Conn, hostID string, env api.E
 		}
 
 	case api.MsgCommandResult:
-		// TODO(P2): persist command.result acks for "did the agent
-		// accept the dispatch?" forensics. Currently the job lifecycle
-		// (job.started → job.finished) is sufficient signal.
-		slog.Debug("ws msg not yet handled", "type", env.Type, "host_id", hostID)
+		var p api.CommandResultPayload
+		if err := env.UnmarshalPayload(&p); err != nil {
+			slog.Warn("ws: decode command result", "host_id", hostID, "err", err)
+			break
+		}
+		// A cancel for a job unknown to the current agent process can never
+		// produce job.finished. Terminalise the orphan server-side.
+		if !p.Accepted && p.Error == "job_not_found" && p.JobID != "" {
+			job, err := deps.Store.GetJob(ctx, p.JobID)
+			if err != nil || job.HostID != hostID {
+				slog.Warn("ws: reject orphan result for foreign or missing job", "job_id", p.JobID, "host_id", hostID)
+				break
+			}
+			if job.Status == string(api.JobQueued) || job.Status == string(api.JobRunning) {
+				if err := deps.Store.MarkJobFinished(ctx, p.JobID, string(api.JobCancelled), -1, nil,
+					"agent reported job not found during cancellation", time.Now().UTC()); err != nil {
+					slog.Warn("ws: terminalise orphaned job", "job_id", p.JobID, "err", err)
+				}
+			}
+		}
 
 	case api.MsgTreeListResult:
 		// Reply to a synchronous tree.list RPC. Route to the waiter