Perform single-qubit operations without calling into Aer

[garrison]

The apply_one_qubit_gate_inplace function can be implemented using numpy because single-qubit operations don't require one to compress the MPS. If we do this, we should be able to avoid some copying and improve performance.