handle openapi bytes (davinci, babbage, curie, ada, etc.)

fix_openai_bytes.py

def try_strip_bytes_prefix(token: str) -> str:
    if token.startswith("bytes:") and "\\" in token:
        return token[6:]
    return token


def try_bytes_fix(tokens: list[str]) -> str:
    # this encode/decode dance fixes places where openai provides
    # "bytes: \\xf0\\x9f\\x98" + "bytes:\\x83" for tokens and replaces
    # with the appropriate emoji (e.g. "😃")
    return (
        "".join(try_strip_bytes_prefix(token) for token in tokens)
        .encode()
        .decode("unicode-escape")
        .encode("latin-1")
        .decode("utf-8")
    )