Distribution/graph/loop semantic

[LeonidElkin]

Summary

This PR introduces a richer characteristic-computation graph and simplifies distribution initialization.

What Changed

• Added support for hyperedges in the registry graph (many -> one), enabling computations that depend on multiple source characteristics.
• Added analytical self-loops in graph views, including labeled loop variants and explicit analytical edge metadata.
• Refactored distribution analytical handling to support multiple analytical computations per characteristic (labeled variants).
• Introduced a shared Distribution.__init__ to reduce boilerplate across distribution implementations.
• Updated .gitignore to exclude local AI-related artifacts. This is a small quality of life change, as many actively use artificial intelligence, but you don't want to see its artifacts in a remote repository.