The last two points are not quite right:

- on outliers: in settings where normalisation makes sense, you cannot leave outliers in your input dataset as they would massively distort your model estimation. If you don't normalise, you would still have to winsorize. In fact, it is better to winsorize even before you normalise so that you do not distort mean, min, max and standard deviation estimates.

- on unit measurement:

you can always un-normalize when you need to go back to the interpretation of your model estimates, so it's very easy to recover back the units of measurement.