Missing continuous outcomes under covariate dependent missingness in cluster randomised trials