Pyrddlgym-jax

New features:
- Added optional parallel evaluation feature to run planner on multiple parameter settings in parallel
- Added basic enum-valued action support in planner
- Added optional hard restarts in planner when it stalls
- Display jax runtime error codes during planner optimization for better debugging

Fixes:
- Improved messages and warnings in planner
- Fix rare casting bug in planner's evaluation code in replanning mode
- Fix replanning mode where jit compilation takes up all the compute budget on the first epoch

**Full Changelog**: https://github.com/pyrddlgym-project/pyRDDLGym-jax/compare/v2.3...v2.4

Changes:
- Added negative binomial approximation
- Added much better poisson and binomial logic for large rate/trials with gradient support in all arguments
- Replace binomial approximation with gumbel-softmax

**Full Changelog**: https://github.com/pyrddlgym-project/pyRDDLGym-jax/compare/v2.2...v2.3

Changes:
- Finally fixed tqdm progress flicker
- Multiple PGPE improvements (entropy regularization, kl constraint)

**Full Changelog**: https://github.com/pyrddlgym-project/pyRDDLGym-jax/compare/v2.1...v2.2

Changes:
- added mean-std and mean-semideviation utilities
- cleanup and better numerical robustness of logic-related code
- added differentiable relaxations for discrete probability distributions (Poisson, Binomial, Bernoulli)
- tuning launch script has dashboard option
- progress bar fixes
- various performance enhancements

**Full Changelog**: https://github.com/pyrddlgym-project/pyRDDLGym-jax/compare/v2.0...v2.1

What's Changed
* Parameter-Exploring Policy Gradient Hybridization (Beta) by mike-gimelfarb in https://github.com/pyrddlgym-project/pyRDDLGym-jax/pull/12


**Full Changelog**: https://github.com/pyrddlgym-project/pyRDDLGym-jax/compare/v1.3...v2.0

Changes:
- Fix constraint handling in planner
- Performance enhancements
 
**Full Changelog**: https://github.com/pyrddlgym-project/pyRDDLGym-jax/compare/v1.2...v1.3