Probe recovery: an answered status probe must confirm the task, not the daemon

The exceptional probe treated any /status answer as 'task keeps
running' - but an answer only proves the daemon lives. A daemon
that crashed and restarted answers from its idle loop (task_id
null, or another task it won since) while the web keeps the
orphaned session as the busy record forever.

The probe now reads the daemon's reported task_id: when it is not
the probed task, the orphan requeues through the same guarded
flip + system comment + release_task bookkeeping as the
unreachable path (the comment names the restart), and the saddle
stays as-is - stamping it offline would only block the requeued
task's next claim. A matching task_id (working, blocked or
suspended) keeps running as before.

src/Services/SaddleService.vala

@@ -1564,6 +1564,35 @@ namespace ClankerSaddle.Services {
             try {
                 var status = yield request_saddle_json(saddle.host, "GET", "/status", null);
                 yield record_status_probe(saddle, status);
+                // An answer proves the DAEMON lives - not that it is
+                // running THIS task. A daemon that crashed and
+                // restarted answers from its idle loop (task_id null,
+                // or another task it won since): the stale session is
+                // a crash orphan, and its task requeues through the
+                // same guarded bookkeeping as the unreachable path
+                // (the comment names the restart). The saddle itself
+                // stays as-is - it is demonstrably alive and asking
+                // for work; stamping it offline would only block the
+                // requeued task's next claim until another ask.
+                var answered_task = status.has_member("task_id")
+                    && status.get_member("task_id").get_node_type() == Json.NodeType.VALUE
+                    && status.get_member("task_id").get_value_type() == typeof(int64)
+                    ? status.get_int_member("task_id") : 0;
+                if (answered_task != running_task_id) {
+                    var flip = connection.create_command(
+                        "UPDATE tasks SET status = 'queued' WHERE id = :id"
+                        + " AND status IN ('assigned', 'working')");
+                    flip.with_parameter<Invercargill.Element>("id",
+                        elem<int64?>(running_task_id));
+                    if ((yield flip.execute_non_query_async()) > 0) {
+                        yield tasks.add_system_comment(running_task_id,
+                            @"Saddle '$(saddle.name)' restarted mid-run (its status probe answered without the task); task requeued");
+                        yield tasks.release_task(running_task_id, OUTCOME_STOPPED, false);
+                    }
+                    yield pulse.bump(0, "crash orphan requeued at probe");
+                    print(@"[ClankerSaddle] Stale saddle '$(saddle.name)' answered its probe without task $(running_task_id); requeued\n");
+                    return;
+                }
                 print(@"[ClankerSaddle] Stale saddle '$(saddle.name)' answered its exceptional status probe; task $(running_task_id) keeps running\n");
             } catch (Error e) {
                 warning(@"[ClankerSaddle] Saddle '$(saddle.name)' sent neither chunks nor results for $(stale_after_seconds())s and its status probe failed ($(e.message)); marking offline");