Regression residuals are correlated and heteroscedastic, again

In a multiple linear regression set up under the assumption that the random errors $\varepsilon_1\dots\varepsilon_n$ are uncorrelated and homoscedastic the residuals $e_1\dots e_n$ are: