[BUG] Replace deprecated batched_dot with pt.sum in KroneckerNormal

- Fixes Issue #7878
- Replace pt.batched_dot(sqrt_quad.T, sqrt_quad.T) with pt.sum(sqrt_quad.T ** 2, axis=-1)
- Computes squared norm per sample using modern PyTensor operations
- Eliminates deprecation warnings and ensures future compatibility

Author: atheendre130505

pymc/distributions/multivariate.py

@@ -2124,8 +2124,8 @@ def logp(value, rng, size, mu, sigma, *covs):
         sqrt_quad = sqrt_quad / pt.sqrt(eigs[:, None])
         logdet = pt.sum(pt.log(eigs))
 
-        # Square each sample
-        quad = pt.batched_dot(sqrt_quad.T, sqrt_quad.T)
+        # Square each sample - compute squared norm for each sample
+        quad = pt.sum(sqrt_quad.T ** 2, axis=-1)
         if onedim:
             quad = quad[0]