fix(prober-queue): time out scans from job enqueue time, not created_at

First live scans false-failed ('scan timed out', completed:0): backlog rows
sat 'queued' ~68 min before the sweep enqueued them, and the reconcile timed
them out immediately because it measured age from created_at. Measure from
job.timestamp so freshly-enqueued jobs get the full window.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: ralyodio

lib/prober-queue.ts

@@ -151,10 +151,10 @@ export async function processDuePortScans(
 
   for (const scan of (runningRows ?? []) as ScanRow[]) {
     const job = (await q.getJob(scan.id)) as Job | undefined;
-    const age = Date.now() - new Date(scan.created_at).getTime();
 
     if (!job) {
-      if (age > RUNNING_TIMEOUT_MS) {
+      // No job and an old row => the job was lost (not just slow to enqueue).
+      if (Date.now() - new Date(scan.created_at).getTime() > RUNNING_TIMEOUT_MS) {
         await markFailed(supabase, scan.id, "prober job missing");
         failed++;
       }
@@ -168,7 +168,10 @@ export async function processDuePortScans(
     } else if (state === "failed") {
       await markFailed(supabase, scan.id, job.failedReason ?? "prober job failed");
       failed++;
-    } else if (age > RUNNING_TIMEOUT_MS) {
+    } else if (Date.now() - job.timestamp > RUNNING_TIMEOUT_MS) {
+      // Measure the timeout from when the job was ENQUEUED (job.timestamp), not
+      // when the scan row was created — a row can sit `queued` a long time
+      // before a job exists, and timing out from created_at kills fresh jobs.
       await markFailed(supabase, scan.id, "scan timed out");
       failed++;
     }