Right-size XLSX ZeroGPU reservation

Set the XLSX inference reservation to 60 seconds based on the measured 52-second deployed run, allowing it to run within the account's remaining ZeroGPU quota. The basic workflow reservation is unchanged.

app.py

@@ -162,7 +162,7 @@ def generate(
     return completion["choices"][0]["message"]["content"].rsplit("</think>", 1)[-1].strip()
 
 
-@spaces.GPU(duration=120)
+@spaces.GPU(duration=60)
 def generate_xlsx_table(
     system_prompt: str,
     user_prompt: str,