Clebsch-Gordan correlate_tensors method

[jwa7]

Following on from PR #237, and completes a TODO item in #239 , the next public API function implemented is the correlate_tensors method. This takes two arbitrary TensorMaps and performs a single Clebsch-Gordan tensor product. The function correlate_density previously implemented in #237, is a special case of this method, used to correlate a density descriptor with itself.

As this is the general case, the method cannot take computational shortcuts as in correlate_density. As such, the public API function signature is simplified to:

def correlate_tensors(
    tensor_1: TensorMap,
    tensor_2: TensorMap,
    selected_keys: Optional[Union[Labels, List[Labels]]] = None,
) -> TensorMap:

where arg selected_keys controls the output of angular and parity channels, and therefore the block combinations performed.

Other Changes

• Unit tests sped up by reducing parametrization and lowering max angular/radial channels of input hypers re @Luthaf comment in Future directions for CG iterations #239
• Breaking change to correlate_density: TensorMaps are now returned with the "order_nu" and "inversion_sigma" key dimensions, even if they are not sparse (i.e. redundant). This metadata should be carried around to allow passing of TensorMaps to CG functions that require it (i.e. correlate_tensors), and can always be easily removed by the user if they wish. Once TensorMaps can be tagged with global (i.e. non-sparse) metadata, they will be moved there (another breaking change, for the future)

[github-actions]

Here is a pre-built version of the code in this pull request: wheels.zip, you can install it locally by unzipping wheels.zip and using pip to install the file matching your system