fix: exit poll loop 10 min after first batch job finishes

Instead of waiting 90 min for all 3 jobs, exit once any job
finishes + 10 min grace period for the others. Harvest with
whatever completed — _SkippedJob wraps the rest.

Typical flow: current finishes at ~10 min, grace gives baseline/
true-color until ~20 min, then harvest with partial results.
Much better than blocking 90 min.

Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)

Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>

app/worker.py

@@ -96,9 +96,11 @@ async def process_job(job_id: str, db: Database, registry: IndicatorRegistry) ->
                 logger.warning("Batch submit failed for %s, will use fallback: %s", indicator_id, exc)
                 jobs = None
 
-            # Poll
+            # Poll — exit early once first job finishes + grace period for others
+            GRACE_PERIOD = 600  # 10 min grace after first job finishes
             if jobs is not None:
                 poll_start = time.monotonic()
+                first_finished_at = None
                 finished = False
                 while True:
                     elapsed = time.monotonic() - poll_start
@@ -114,6 +116,17 @@ async def process_job(job_id: str, db: Database, registry: IndicatorRegistry) ->
                         logger.warning("Batch job failed for %s: %s", indicator_id, statuses)
                         break
 
+                    # Track when first job finishes
+                    if first_finished_at is None and any(s == "finished" for s in statuses):
+                        first_finished_at = time.monotonic()
+                        print(f"[Aperture] {indicator_id}: first job finished, {GRACE_PERIOD}s grace for remaining")
+
+                    # Grace period: once any job finished, give others 10 min then harvest partial
+                    if first_finished_at and (time.monotonic() - first_finished_at) >= GRACE_PERIOD:
+                        logger.info("Grace period expired for %s, harvesting partial results", indicator_id)
+                        print(f"[Aperture] {indicator_id}: grace period expired, proceeding with partial results")
+                        break
+
                     if elapsed >= BATCH_TIMEOUT:
                         logger.warning("Batch poll timeout after %.0fs for %s", elapsed, indicator_id)
                         break