Eval Charts add animated, accessible layer

This is the kind of utility layer that looks narrow until teams start using evals to compare real model behavior in front of humans; then the UX becomes part of the product. Provider logos and richer tooltips make model comparisons easier to scan, especially when multiple vendors or checkpoints are on the same chart. Touch support and a11y are the tell that this is meant for real dashboard usage, not a one-off demo screenshot. Building it from scratch suggests general-purpose chart libs were too generic for eval-specific storytelling and interaction. If it ships open source, it could become a useful UI primitive for LLM eval dashboards and internal benchmark tooling.