Noticed while working on: #31701

_check_sample_weight gives the following error message:

I don't think sample weights can ever be a scalar. Indeed later we check:

So it actually can't be a scalar.

Just to be sure I searched for "weight :" and "weights :" in our docstrings and found that we always ask for and ndarray for sample_weights.

For clustering weights is a string or callable. In the times where weights can be a scalar (e.g., in VotingRegressor or 'class_weights', we never use _check_sample_weight.)