Keep watching the job through transient fetch failures; don't freeze the screen

reprocessInPlace's poll loop swallowed fetchJobs failures as an empty
list and broke on the resulting missing status, so ONE dropped poll
during a 10-minute summarize ended the watch early: chips stayed
disabled, the screen showed stale content, and only leaving to the
library and back (fresh openDetail fetch) displayed the finished
summary — the exact reported symptom.

Now only a successful fetch reporting a terminal state ends the loop;
transient failures keep polling for ~30s, and a genuinely-lost engine
says so in plain words instead of silently stranding the screen. A
missing job row on healthy fetches gets 3 grace cycles before giving
up (engine rewrites rows mid-requeue).

app/src/main/kotlin/com/shonar/desktop/DesktopState.kt

@@ -1632,11 +1632,30 @@ class DesktopState(private val appDir: File = defaultAppDir()) {
                     return@launch
                 }
             }
-            // Poll the target job to a terminal state.
+            // Poll the target job to a terminal state. A job this long
+            // (10+ min = hundreds of polls) WILL hit a transient fetch
+            // failure; treating one as terminal (the old `orEmpty()` →
+            // st == null → break) froze the screen on stale content until
+            // the user left and returned — exactly the reported symptom.
+            // So: only a SUCCESSFUL fetch reporting a terminal state
+            // breaks; failures keep polling up to ~30s of contact loss.
+            var miss = 0
+            var gone = 0
             while (true) {
                 delay(2000)
-                val jobs = runCatching { parseJobs(provider.fetchJobs(remoteId)) }.getOrNull()
+                val fetched = runCatching { parseJobs(provider.fetchJobs(remoteId)) }
-                    .orEmpty()
+                if (fetched.isFailure) {
+                    if (++miss >= 15) {
+                        d0.error = "Lost contact with the engine while " +
+                            "processing — reopen this recording to check."
+                        d0.busy = null
+                        _detail.value = d0.copy()
+                        return@launch
+                    }
+                    continue
+                }
+                miss = 0
+                val jobs = fetched.getOrDefault(emptyList())
                 d0.jobs = jobs
                 jobLabel(jobs)?.let { d0.busy = it }
                 // Mirror live progress onto the library row too — otherwise
@@ -1646,7 +1665,10 @@ class DesktopState(private val appDir: File = defaultAppDir()) {
                 rescanStatuses()
                 _detail.value = d0.copy()
                 val st = jobs.firstOrNull { it.jobType == job }?.status
-                if (st == null || st in TERMINAL) break
+                if (st in TERMINAL) break
+                // Job row missing from a healthy fetch: give it a few
+                // cycles (the engine rewrites rows mid-requeue) then stop.
+                if (st == null && ++gone >= 3) break
             }
             setLive(d0.file.name, null)
             rescanStatuses()