Cap GPU duration reservation at 140s to clear free-tier per-call limit

get_inference_duration() returned a padded worst-case that, once reserved,
exceeded free-tier ZeroGPU's ~300s per-call cap -> 'The requested GPU duration
is larger than the maximum allowed'. Cap the reservation at 140s, which still
covers real p90 (~107s) and the 150s caller timeout, so generations run at any
length.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M4Dr1cMq3pUbTUnXXRs9V1

app.py

@@ -441,6 +441,11 @@ def get_inference_duration(
     if safe_mode:
         total_time = total_time * 1.30
 
+    # Cap reserved GPU duration so the reported reservation (×2 for xlarge) stays
+    # under free-tier ZeroGPU's ~300s per-call limit — the padded worst-case was
+    # tripping "requested GPU duration larger than the maximum allowed". 140 →
+    # ~280s reported, still covers real p90 (~107s) and the 150s caller timeout.
+    total_time = min(total_time, 140)
     return total_time