Function read_request_with 

pub async fn read_request_with<S>(
    s: &mut S,
    role: &str,
    limits: &ProxyLimits,
    bearer: Option<&[u8]>,
) -> Result<ProxyRequest, ProxyError>
where
    S: AsyncRead + AsyncWrite + Unpin,

read_request for a listener with a bearer token (an exposed listener, see InferProxy::start_exposed). The token is checked right after the head, before any body is read, and then removed from the request: it is this listener’s credential, not the model server’s. With a token the Host need not be loopback (the credential, not the address, is the guard); without one it must be.