Split computing of PT2 and Selection to chunck for external

[TApplencourt]

The external space should be generated by chunk
And then computing the PT2 and selection in this chunk

This should avoid OOM problem

[seanr7]

Update:

Current `naive' approach (in the code): Connected determinants are generated and then split over ranks. This is a working solution for small problems, but will not scale.

New `constraint-based' approach (to be completed): From https://arxiv.org/abs/2303.05688, https://arxiv.org/abs/1808.02049; connected determinants are generated by constraint (e.g., their three highest occupied alpha spin orbitals), and constraints are dispatched across ranks. This uniquely determines a disjoint partitioning of the connected space.

To do:

1. Implement the algorithm for generating connected determinants based on constraints (Algorithm 3 in https://arxiv.org/abs/1808.02049). (Done as of 6-9, see: https://github.com/seanr7/QuantumEnvelope/tree/new_selection)
2. Implement the load-balancing algorithm for constraints (Algorithm 3 in https://arxiv.org/abs/2303.05688)
3. Implement the distributed selection algorithm for PT2 (Algorithm 4 in https://arxiv.org/abs/2303.05688)

[seanr7]

I think we can close this issue and do a new PR with the recent changes.