The blocks result only depends on tensor itself, does not depend on the
auxiliary blocks object. The former relation has been described by the base
argument in array_t constructor.

This function is useful when spliting a tensor to two sqrt tensor especially for
fermionic case.

Since sqrt for the tensor contains negative number is meaningless.

Update the openblas version to 0.3.21 in the build script used by github action
for the windows platform.

add parameter `use_natural_gradient_by_direct_pseudo_inverse`,
`natural_gradient_r_pinv` and `natural_gradient_a_pinv` for high/mid-level API
in `gm_run` to use direct pseudo inverse to calculate natural gradient.

Method detail could be found at arXiv2302.01941.

Add hamiltonian to another part of density matrix for Heisenberg/Hubbard/tJ
model. It ensures the result density is unitary despite of errors introduced
by contract and approximation.

This control whether the hamiltonian should apply to single side or both side of
density matrix.

To avoid copying from these files and causing bug in the future when dealing with
complex model.

…s state.

The abstract_lattice for density matrix state also accept side argument.

Add import importlib.util/machinery since sometimes they are not
imported.

Move `Configuration` for ap state to an individual package named
`tetraux` from `TAT.py`, which is not related to the tensor itself.