fix: start scheduler at module load so gunicorn workers run the scout

Previously _scheduler_engine.start() was inside if __name__ == '__main__':
which gunicorn never enters (it imports the module as 'app', not '__main__').
The scheduler was dead on every HuggingFace deployment — no new signals.

Changes:
- app.py: move _scheduler_engine.start() + atexit.register(stop) to
module level, immediately after app.layout = _layout(). Runs under
both gunicorn and direct python execution. Remove duplicate start()
from __main__ block.
- core/scheduler.py: set next_run_time=datetime.now(timezone.utc) so
the first scout cycle fires immediately on startup instead of waiting
6 hours. Subsequent runs still follow the 6-hour interval.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

app.py

@@ -19,6 +19,7 @@ Sponsor requirements implemented:
 
 from __future__ import annotations
 
+import atexit
 import hashlib
 import os
 import sys
@@ -1644,6 +1645,14 @@ def _layout() -> html.Div:
 
 app.layout = _layout()
 
+# ── Start scheduler at module load (works under gunicorn and __main__) ──
+# Must be AFTER app.layout so Dash is fully initialised.
+# gunicorn imports this module directly — __main__ guard would skip it.
+_scheduler_engine.start()
+atexit.register(_scheduler_engine.stop)
+log.info("SchedulerEngine started at module load. Next scout: %s",
+         HEALTH.get("next_run_utc"))
+
 
 # ─────────────────────────────────────────────────────────────
 # Callbacks
@@ -2082,7 +2091,7 @@ def lens_search(topic: str | None, custom: str | None, _ns: int) -> html.Div:
 if __name__ == "__main__":
     _preflight()
 
-    _scheduler_engine.start()
+    # Scheduler already started at module load — just log status
     stats = _db_stats_cached()
     log.info("App starting — Astra DB: %d signals, HuggingFace: %s",
              stats["total"], "OK" if _HF_OK else "NO KEY")
@@ -2096,4 +2105,3 @@ if __name__ == "__main__":
     # Use PORT env var (Hugging Face / Render) or default to 7860
     port = int(os.environ.get("PORT", 7860))
     app.run(debug=False, host="0.0.0.0", port=port)
-    _scheduler_engine.stop()

core/scheduler.py

@@ -186,14 +186,16 @@ class SchedulerEngine:
             timezone="UTC",
         )
 
-        # Scout job — runs every N hours, NOT immediately on startup
+        # Scout job — fires immediately on first start, then every N hours.
+        # next_run_time=now() ensures the first cycle runs right away so the
+        # dashboard is populated without waiting 6 hours after deployment.
         self._scheduler.add_job(
             _run_scout_cycle,
             trigger=IntervalTrigger(hours=self._interval_hours),
             id="scout",
             name="PESTEL Scout Cycle",
             replace_existing=True,
-            next_run_time=None,
+            next_run_time=datetime.now(timezone.utc),
         )
 
         # Heartbeat — every 30 s