TODO: implementation for any axis is pretty straightforward (pad_shapeleft +
DimShuffle)

"missing storage error"

* clearer error messages when input data does not match either indices or kmap

test file.
+ misc. tweaks

of [False]*<dim>

@TODO: this whole kernel map thing should've probably been implemented as its
own op, which would sit between the kernel taps and the CSM op (a "Shuffle" or "TensorView" op). Revisit at a later date

* bug fix: force todense of sparse matrices in check_equal, as two sparse
matrices are never equal in scipy 0.6
* numeric_grad no longer cares that all parameters are of the same type
* new Reshape optimization (@todo: reshape-flatten optimization)