Raise ZeroGPU durations to 20s fast, 60s high-detail

10s/30s caused aborts on cold ZeroGPU instances where Triton compilation
adds 10-20s to the first request. Each new concurrent instance recompiles,
so this isn't a one-time startup cost.

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

app.py

@@ -352,7 +352,7 @@ def infer(images_b64_json, prompt, seed, randomize_seed, guidance_scale, steps,
         raise
 
 
-@spaces.GPU(duration=lambda *a, **kw: 10 if (len(a) > 7 and a[7] == "fast") else 30)
+@spaces.GPU(duration=lambda *a, **kw: 20 if (len(a) > 7 and a[7] == "fast") else 60)
 def _infer_gpu(pil_images, prompt, seed, guidance_scale, steps, width, height, mode):
     _cuda_ok = torch.cuda.is_available()
     timer = _InferTimer(_cuda_ok)