vLLM using built-in hash() from Python 3.12 leads to predictable hash collisions in vLLM prefix cache

...