refactor of group 2 (gp): improve GP numerical stability and JIT compatibility

- Fix latent bug in `gp_predict` where non-static `log_func` arguments
  caused TypeError during JAX JIT compilation.
- Implement a wrapper for `gp_predict` to handle shape validation and
  logging outside of the JIT-compiled core.
- Update `gp_predict` to fail fast with `ValueError` on `dy_sq` length
  mismatches instead of silently falling back to a default diagonal.
- Update `_configure_jax_precision` to automatically enable float64
  when GP components are detected in the model.
- Add a "do-it-at-your-own-risk" warning when float32 is manually
  forced despite the presence of precision-critical components (GPs or
  ref_delta parameters).
- Clean up dead code, unused imports, and missing module documentation
  in `gp.py` and `kernels/factory.py`.