ORMAS implementation

[kgasperich]

This allows applying constraints to the determinants in the wavefunction using the occupation-restricted multiple active space (ORMAS) method.
The multiple active spaces are specified by the first orbital index in each space: bitmasks.ormas_mstart, which is a 1d array of length bitmasks.ormas_n_space.

For example: in a system with 50 orbitals, if bitmasks.ormas_mstart is set to (1,10,15,30), there will be four active spaces, with the first space consisting of orbitals [1,2,...,9] the second space [10,11,...,14], the third space [15,16,...,29], and the last space [30,31,...,50].
If the first starting index is not set to 1, then there will be no restriction on the orbitals with indices below the start of the first ORMAS active space.

The occupation restrictions are specified by two 1d arrays of length bitmasks.ormas_n_space: bitmasks.ormas_{min,max}_e, which are the minimum and maximum allowable number of electrons within the orbitals in that active space.

ormas_bitmask contains bitmasks corresponding to each space for easier checking to see if a determinant is allowed.
The subroutine ormas_occ(key_in, occupancies) will set occupancies (list of ints) to the number of electrons in each ORMAS space for the determinant key_in. If each of these is between the allowable ormas_{min,max}_e, then the determinant is allowed.
The function det_allowed_ormas(key_in) returns .T. or .F. depending on whether the determinant is allowed (it will exit early with .F. as soon as it finds a space that does not obey the min/max elec constraints).

The actual filtering is performed during the selection in the same place as other similar restrictions (CAS, seniority, DDCI, etc.)

As it's currently written, this PR only allows ORMAS spaces which are contiguous; it should not be difficult to modify it so that each space is given as a list of orbitals (rather than just a starting index for a space that implicitly extends up to the MO preceding the start of the next space).

This implementation also only allows disjoint active spaces. This could be changed, but it would complicate things like ORMAS-SCF or doing rotations to form natural orbitals where the user might not want to mix orbitals between different active spaces. For cases where overlapping spaces would be necessary, it should be possible to specify an equivalent set of restrictions using disjoint spaces with a more complicated set of allowable electron counts in each space. Another workaround would be to allow overlapping spaces, but when any orbital rotations are performed each intersection of spaces will be considered its own space (i.e. not allowed to mix with any of the overlapping sets that it is a part of).

[kgasperich]

Maybe it would be cleaner and more consistent if this were separated in the same way as the casscf is?
The current implementation is just an additional level of independent filtering, so it would be possible to call qp_run fci and get something that is an ORMAS-CI (if do_ormas == .T.) rather than a true FCI.

[scemama]

Thanks @kgasperich !
I think it is better like that than making a special module out of it, because then you can naturally do the intersection between ormas and FCI, CISD, selected or not, etc.

[scemama]

Looks good to me!