THC TDA

[mkakcl]

• Adds thc.py introducing the new function of completing TDA calculations using THC integrals. Using this method should scale as N^3 in comparison to the previous Cholesky decomposition method (scales as N^4)
• Current implementation requires the THC tensors to be calculated from a different source and then imported into the code. This is currently achieved within the ints.py. Future implementations will hopefully directly interface with a method to calculate THC ERIs.

[obackhouse]

Plan after call:

• thc_opts only additional keyword argument in BaseGW, for now only contains filepath
• For now, just have momentGW/thc.py that contains a new Integrals and TDA class that inherit from the standard versions
• Specify THC via polarizability i.e. polarizability="thc-dtda", which controls what is returned by GW.ao2mo and GW.build_se_moments
• Move the fock.py Fock matrix building functions inside the Integrals classes - this allows us to also remove the calls to PySCF in build_se_static in place of these functions, so we can maintain scaling with THC
• Ignore self-consistency, MPI for now

I think this should make everything interface nicely without chaning much existing code

[obackhouse]

I've changed some stuff around, made variable names more consistent with the DF implementation, and I think things are looking good. Still need tests, and to stress test the miscellaneous features in the GW solver. Then self-consistency and parallelism.

Not sure what our merge policy here is - perhaps this should just live in a development branch until we have a usable interface with the THC integral generation.

[obackhouse]

Can we move the MPI stuff to a new branch? Then we can look to getting this PR merged this week, I think it's ready to go

[mkakcl]

Branched MPI changes and reverted to pre MPI

[obackhouse]

on second thought, maybe we want to get the self-consistency working before merging

[obackhouse]

LGTM, great stuff @mkakcl