This removes the need for a special 'ndarray' case, and should generalize to
Python and numpy numeric types, as well as CudaNdarray, and even other types.

Also test it.
This should fix the problem experienced by Justin Bayer.

Remerged 3446d3f07946 and 0fd23055db15: the merge in 491d6aa69509 was incorrect due to badly configured merge tool

* compute_test_value=True enables on-the-fly execution of perform method
* the user provides default test values through the tag scratchpad, i.e.
T.dmatrix().tag.test_value = numpy.random.rand(4,6)