Add concentration-space LOO and plotting tools

Introduce a full concentration-space LOO workflow and visualization support.

- Add catalax.mcmc.loo: reconstruction of per-draw concentration log-likelihoods for mechanistic and surrogate (rate-based) fits, build_loo_idata, loo(), loo_pointwise(), compare(), and consistency_check(); supports Euler integration variants and options for using measurement errors or reusing sampled noise.
- Export LOO functions and new plotting hooks (plot_loo_influence, plot_loo_heatmap) from catalax.mcmc.__init__.
- Add plotting utilities in catalax.mcmc.plotting for visualizing per-point Pareto-k / elpd influence and heatmaps, plus related helpers and color scales.
- Add convenience methods on HMCResults (loo, compare, loo_pointwise, loo_consistency_check, plot_loo_influence) to call the new LOO API.
- Fix likelihood/discrepancy handling in BayesianModel (use HalfNormal prior for sigma and treat sigma_total as sampled sigma rather than sigma**2) to align noise semantics with reconstruction.

These changes enable consistent LOO comparison across mechanistic and surrogate-mode fits and provide plotting helpers to inspect high-influence observations.

Author: JR-1991

catalax/mcmc/__init__.py

@@ -1,5 +1,22 @@
 from .mcmc import run_mcmc, MCMCConfig, BayesianModel, HMC, HMCResults
-from .plotting import plot_corner, plot_posterior, plot_trace, plot_forest, summary
+from .loo import (
+    loo,
+    compare,
+    consistency_check,
+    build_loo_idata,
+    reconstruct_log_likelihood,
+    loo_pointwise,
+    LooPointwise,
+)
+from .plotting import (
+    plot_corner,
+    plot_posterior,
+    plot_trace,
+    plot_forest,
+    summary,
+    plot_loo_influence,
+    plot_loo_heatmap,
+)
 from .protocols import (
     PreModel,
     PostModel,
@@ -35,4 +52,13 @@
     "PostModelContext",
     "pre_model",
     "post_model",
+    "loo",
+    "compare",
+    "consistency_check",
+    "build_loo_idata",
+    "reconstruct_log_likelihood",
+    "loo_pointwise",
+    "LooPointwise",
+    "plot_loo_influence",
+    "plot_loo_heatmap",
 ]