Multiplying kernels with Python floats results in scaled kernel objects.
But multiplication of kernels with a scalar numpy floats results in an ndarray objects.

>>> import numpy as np
>>> from astropy.convolution import Gaussian2DKernel
>>> psf = np.float64(42) * Gaussian2DKernel(3)
>>> print(type(psf))
<type 'numpy.ndarray'>
>>> psf = 42 * Gaussian2DKernel(3)
>>> print(type(psf))
<class 'astropy.convolution.kernels.Gaussian2DKernel'>

I think this behaviour is confusing (took me 15 minutes to pin down and understand in my script) and should be changed so that numpy scalar float times kernel is again a kernel.

cc @astrofrog @adonath Do you agree? Can you fix this?