Trim response to active qubits/variables only

[randomir]

Currently result/samples are expanded to full length of working qubits (2048) even when user has only a couple of active ones.

[randomir]

It's easy to trim the trailing inactive qubits, but if we're using a list for the response, "holes" have to filled out with (perhaps) NaNs. The alternative is to, by default, use dict/mapping for the response, and only if user requested numpy_matrix, return the complete matrix with NaNs inplace of inactive qubits. In that case, make sure active_variables is exposed and it can easily be used for selecting only the active ones from the response matrix.