Full benefit from minimal representation

Currently, the wave functions are stored in the minimal representation. However, when transformed to sample values with I(), as occurs in getE() and H(), the output lives in the full FFT box and consumes an unnecessarily large amount of memory (sixteen times more than is needed), whose allocation and deallocation also tends to slow the calculation. To avoid this, we should apply I() only to one column of the wave functions at a time. To achieve this, make the following modifications.