fix: honor ENABLE_KVCACHED in autopatch enable check

[cui36]

To solve the issue #320, #317 and #316, by setting kvcached using unified arg ENABLE_KVCACHED.

[jiarong0907]

def _env_enabled() -> bool:
    return os.getenv("KVCACHED_AUTOPATCH", "false").lower() in ("true", "1")
    return (
        os.getenv("ENABLE_KVCACHED", "false").lower() in ("true", "1")
        or os.getenv("KVCACHED_AUTOPATCH", "false").lower() in ("true", "1")
    )

Not sure this is a good design because enabling kvcached is separated from enabling autopath. I can imagine someone manually integrates kvcached with vLLM and thus doesn't need our autopatching.