Parax v0.7 sharpens JAX modeling API

This is a niche but solid refinement of a useful idea: make JAX modeling more parameter-centric without forcing users into a heavyweight object system.

• –The big value is ergonomic modeling of constraints, priors, and metadata, which is exactly the kind of boilerplate that tends to spread across scientific JAX codebases
• –Its Equinox interoperability matters more than it sounds like, because it lets teams keep functional PyTree workflows while still getting structure where they need it
• –The new bounded optimization and Bayesian sampling examples are the right proof points; those are real downstream use cases, not toy demos
• –This is still a specialist library, so it reads more like infrastructure for advanced JAX users than a broad AI tooling breakout