Allow greater control over the dimension of the SCI subspace

[kevinsung]

What should we add?

The solve_fermion function accepts an array of bitstrings, where each bitstring has an "alpha" half and a "beta" half. In order to form the actual SCI subspace, these bitstrings are split into alpha and beta halves (this is done in bitstring_matrix_to_ci_strs) and the actual subspace is formed as the cross product of the sets of unique alpha strings and unique beta strings.

This procedure makes it difficult to directly control the subspace dimension, which is a useful feature. I'm not sure what we should do. Perhaps in hindsight, we should not have deprecated the option to pass the CI strings directly as input.