Bisect AOTI compile duration upward: 600 → 1000

600s passed the cap check (boot died later on NVML flake). Probing higher
to find the actual ceiling — cap is in [600, 1200).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

optimization.py

@@ -49,7 +49,7 @@ INDUCTOR_CONFIGS = {
 
 def optimize_pipeline_(pipeline: Callable[P, Any], *args: P.args, **kwargs: P.kwargs):
 
-    @spaces.GPU(duration=600)
+    @spaces.GPU(duration=1000)
     def compile_transformer():
 
         # Run warmup pass to capture transformer inputs (needs all components on GPU)