Add variable size batched output support for TransformDPP in fk::executeOperations

[morousg]

We want to add in fk::executeOperations (CPU Host code) the analisys of both the Read and Write Operations, to decide how to handle the active threads in the TransformDPP.

If the Batched output contains pointers of different sizes and shapes, we will need to diverge the x,y threads according to the shape on each z plane.

Otherwise, if the output is a Tensor or a batch of same size/shape pointers, then we can use the maximum size for the BatchRead x and y.

Remember that a default or background value, is still a value, and it needs a thread and a memory position to be written.

Therefore:

1. The DPP will be responsible to decide which threads participate.
2. The Read or BatchRead Operation, will decide which value each active thread returns, using the width and height of the image or a ROI, which can be equal or smaller than the active threads.