Make remaining message and prior xp paths JAX-traceable

[Jammy2211]

Summary

Make the remaining message and compound-prior array-namespace paths genuinely JAX-traceable.

• select the JAX backend while constructing traced Beta/Gamma messages
• dispatch Beta/Gamma log-partition special functions through SciPy or jax.scipy
• propagate xp recursively through arithmetic and modified priors
• use xp.log / xp.log10 for compound-prior modifiers
• add scalar, batched, nested-prior, and NumPy-parity JIT regressions

API Changes

Prior.instance_for_arguments gains an optional xp=np keyword, aligning direct priors with AbstractPriorModel.instance_for_arguments. Beta/Gamma message construction and compound-prior evaluation now honor JAX inputs instead of coercing them through NumPy. Existing NumPy calls and defaults are unchanged.

See full details below.

Test Plan

• Direct tests against untouched post-Use xp.stack in messages so JAX can trace fixed arrays #1460 main: 12 expected JAX failures, 20 passes
• Direct branch tests on JAX 0.10.2: 32 passed
• Direct branch tests on supported-minimum JAX 0.7.0: 32 passed
• Focused message/prior suites: 317 passed
• Full PyAutoFit suite: 1701 passed, 4 skipped

Full API Changes (for automation & release notes)

Changed Signature

• Prior.instance_for_arguments(arguments, ignore_assertions=False, xp=np) — adds an optional backend argument used by recursive compound-prior evaluation.

Changed Behaviour

• BetaMessage and GammaMessage constructors select JAX broadcasting for traced parameters.
• BetaMessage.log_partition(xp=jnp) uses jax.scipy.special.betaln.
• GammaMessage.log_partition(xp=jnp) uses jax.scipy.special.gammaln and xp.log.
• Arithmetic and modified priors propagate xp to nested children.
• Log and Log10 dispatch through xp.log and xp.log10.

Migration

• None. Existing callers remain valid; xp is optional and defaults to NumPy.

Closes #1459

Generated by the PyAutoLabs agent workflow.