[Feature] Adding blackjax sampling functionality

[yallup]

Adds a JAX-native nested sampling fitter using the blackjax library, enabling end-to-end JAX-based Bayesian inference.

Summary of current changes:

• New Fitter: Introduces blackjax_fit.py, which implements the nested sampling loop using blackjax.nss.
• Generalized Prior: The bayesian_prior function is refactored. It now returns a dictionary containing prior_transform (for UltraNest), and log_prob and sample functions (for BlackJAX).
• New Dependencies: Adds blackjax, tensorflow-probability (for prior distributions), and anesthetic (for results handling). The last two of these are optional but are included into the pyproject.toml for ease of use
• Configuration: The new fitter is integrated into the configuration system, accepting blackjax_settings in runcards.
• Example Runcard: A new runcard, lh_fit_closure_test_blackjax.yaml, is included to demonstrate usage.

Todo

• Correct the returned BayesianFit results to be more compatible with colibri standards
• Test Vectorization of the likelihood, particularly on a GPU
• Add tests

[yallup]

As we are preparing this I would suggest revisiting if we can switch to the tagged version:
https://github.com/handley-lab/blackjax/releases/tag/nested-sampling-beta.1

There are some changes incoming to this main PR branch as we finalise it that may be breaking, so it would be better if possible to point to a stable release so we can manage the (hopefully) finished merged PR integration with colibri at a later date. I think in theory it should be a like for like swap but I think there were some conda env problems when we last looked at this