fix(upload): eagerly cancel upload work when prerequisites disappear

The worker-level gate from the previous commit caught the case
lazily (next scheduled run sees the missing prerequisite and bails),
but the cancelled work still sat in WorkManager's queue until its
backoff timer elapsed. If the retry was parked on the 32m / 1h step,
that's the window the cancelled work would sit "scheduled but doomed".

Now the three opt-out paths drop the unique-work entry immediately:
  - SyncManager.clearSyncFolder       → cancels BOTH workers
  - SettingsViewModel.unlinkOnline    → cancels eucstats
  - SettingsViewModel.deleteOnline    → cancels eucstats (post-success)

clearSyncFolder cancels both because eucstats's rider id file lives in
the SAF folder; losing the folder loses online identity too.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: eried

app/src/main/java/com/eried/eucplanet/data/sync/SyncManager.kt

@@ -376,6 +376,11 @@ class SyncManager @Inject constructor(
             lastSettingsBackupAt = null,
             onlineUploadEnabled = false,  // online upload requires a folder
         ))
+        // Both destinations just lost their prerequisites: folder is gone, and
+        // the rider id file went with it. Drop any pending retries so we don't
+        // sit on a backed-off worker that would only no-op when it finally fires.
+        cancelTripUpload()
+        cancelEucStatsUpload()
     }
 
     /** The chosen folder's DocumentFile, or null if none or no longer accessible. */
@@ -687,6 +692,23 @@ class SyncManager @Inject constructor(
         )
     }
 
+    /** Cancel any queued / in-backoff folder-upload work. Used when the rider
+     *  unlinks the sync folder so a long-tail retry doesn't sit waiting for a
+     *  destination that no longer exists. */
+    fun cancelTripUpload() {
+        WorkManager.getInstance(context).cancelUniqueWork(UPLOAD_WORK_NAME)
+    }
+
+    /** Cancel any queued / in-backoff eucstats-upload work. Used when the rider
+     *  toggles leaderboards off, deletes their account, or unlinks the sync
+     *  folder (since the rider id lives in that folder). Without this, a retry
+     *  parked on the 32m / 1h step would still fire later and find the
+     *  prerequisites gone; cancelling here keeps the trip icons and WorkManager
+     *  state in lockstep. */
+    fun cancelEucStatsUpload() {
+        WorkManager.getInstance(context).cancelUniqueWork(EUCSTATS_UPLOAD_WORK_NAME)
+    }
+
     /** Schedule the eucstats worker for [attempt]. See [scheduleTripUploadAttempt]
      *  for semantics; this one also carries the CONNECTED network constraint. */
     fun scheduleEucStatsUploadAttempt(attempt: Int) {

app/src/main/java/com/eried/eucplanet/ui/settings/SettingsViewModel.kt

@@ -2013,6 +2013,10 @@ class SettingsViewModel @Inject constructor(
                 }
             }
             settingsRepository.update(settingsRepository.get().copy(onlineUploadEnabled = false))
+            // Drop any queued / in-backoff retries; the gate inside the worker
+            // would catch them anyway but cancelling now keeps the unique-work
+            // entry from sitting around until its delay elapses.
+            syncManager.cancelEucStatsUpload()
         }
     }
 
@@ -2085,7 +2089,12 @@ class SettingsViewModel @Inject constructor(
             // Remove the local recovery file too, so the just-deleted profile is
             // not offered for "restore" on the next Join (it no longer exists
             // server-side). Best-effort, only when the server delete succeeded.
-            if (ok) syncManager.deleteRiderIdFile()
+            if (ok) {
+                syncManager.deleteRiderIdFile()
+                // Drop any queued / in-backoff eucstats retries so they don't
+                // wake up later and hit a deleted account.
+                syncManager.cancelEucStatsUpload()
+            }
             onResult(ok)
         }
     }