DOC: Update float32 mean example in docstring

[kjordahl]

Computing the np.mean() of a float32 array is less inaccurate than it used to be (see e.g. #5681). This just updates the docstring to the value returned by the current 1.11 release and dev versions.

This should be caught in a doctest, but there are a lot of other errors for me when I try doing that. I'm probably missing a simpler way to isolate this single doctest.

[charris]

Thanks @kjordahl . If you put [ci skip] in the commit message tests will be skipped for documentation updates like this.