3dfftconv

Don't make the kernels debuggable by default, but add those arguments to the 'nvcc-safe' list.

Conv gemm non-square kernel support

Extend documentation (plus small fix)

Cleanup CUDA convolution tests

Add a blocksparse multiplication implementation

Remove the restriction on indexing a broadcastable dimension.

Update opt to replace Conv3D, ConvGrad3D and ConvTransp3D instead of GpuConv3D, GpuConvGrad3D and GpuConvTransp3D.

Conflicts:
	doc/library/tensor/nnet/conv.txt

@as_op() documentation

This should help older GPUs run at all and newer GPUs fit more blocks
on one SM.

With this change the code is cc 2.0+ compatible.  But it will only be
fast on cc 3.0+ cards (due to atomicAdd).

Also some small kernel speedups elsewhere.

Also address some other issues that came up in code review.

Remove the beta parameter since it's always 1 anyway.