Redesigned metatensor-based Clebsch Gordan iterations

[jwa7]

Closes #178 and #185 . New API design for performing metatensor-based Clebsch Gordan iterations.

This PR:

• Public functions correlate_density and correlate_density_metadata, which perform iterative CG tensor products of a density with itself, returning the a density correlation descriptor of specified correlation_order, or just the metadata that would wrap it.
• Dispatches to numpy, or Mops (if installed). Placeholder for torch dispatch, but this needs fully checking. (@jwa7 @frostedoyster)
• Unit tests (@agoscinski, @curiosity54, @jwa7)
• Docstrings and the .rst from Added cg_utils to Rascaline #185 (@jwa7, @curiosity54, @ecignoni)

Not in this PR (see #239):

• Public function correlate_tensors for doing a single CG product step on 2 arbitrary TensorMaps with the same samples
• Public function correlate_tensors_different_samples for doing a single CG product step on 2 arbitrary TensorMaps with different samples.
• Standalone documentation / tutorials / help guides

Authored by @jwa7 , @agoscinski , @curiosity54, @frostedoyster , @DivyaSuman14 , @Luthaf , @ecignoni

---

📚 Documentation preview 📚: https://rascaline--237.org.readthedocs.build/en/237/

[Luthaf]

small comments, I'll do a full review later

[jwa7]

@Luthaf In order to make the CI happy I had to add some dependencies in the min-deps env - for CG tests we need ase, metatensor-operations and sympy. Of course not ideal as defeats the purpose of being minimal dependencies - what's the best way to get around this? Is there a way to only run CG tests at a higher level of dependencies?

[Luthaf]

Is there a way to only run CG tests at a higher level of dependencies?

yes! https://docs.pytest.org/en/7.1.x/how-to/skipping.html#skipping-on-a-missing-import-dependency

[jwa7]

Thanks for the review @Luthaf ! I've made the changes re: your suggestions :)