Refactor：Support custom HexxR to OperatorEXX

[maki49]

The future implementation of density matrix functional theory will use a different Hexx(R) from the one in LM.

@dyzheng This PR will conflict with #3451, please merge this first and let me solve the conflict in #3451, thanks.

[dyzheng]

I agree with this refactor PR, do you have further plan to add Hexxd/Hexxc into HContainer* hR?

[maki49]

I agree with this refactor PR, do you have further plan to add Hexxd/Hexxc into HContainer* hR?

Not yet, but I did think about it.
We need to use LibRI's container (RI::Tensor) to calculate Hexx(R) at LibRI's interface. To use HContainer, we have to add the transformer between the two types of containers. The advantage then is to reuse the Fourier transformers between R and k. However, since the range of R is different between Hexx(R) (BvK supercell) and other terms (adjacent table), I think the current way (directly using the Fourier transformer as the transformer between containers) is an elegant trick to some extent.