Mirror R's tryCatch fallback around NNS.dep in regression dependence

R Regression.R wraps the NNS.dep call in tryCatch with a 0.1 fallback
before averaging with the copula measure; the port only guarded the
copula half. Wrap the nns_dep call the same way so the order=NULL
dependence chain is a literal transcription of the R block.

Verified against live R NNS 13.1 on nine datasets spanning n<100 and
n>=100: NNS.dep and NNS.copula components bit-identical, composed
dependence within 1 ulp, and every resolved dep.reduced.order equal
(order=NULL and integer overrides).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BPbZvtDw4h2XJo9w5hw57h

Author: claude

src/nns/regression.py

@@ -1071,7 +1071,12 @@ def _as_point_est(point_est: NDArray[np.float64] | float | None) -> NDArray[np.f
 
 
 def _regression_dependence(x: NDArray[np.float64], y: NDArray[np.float64]) -> float:
-    dep = nns_dep(x, y, asym=True)["Dependence"]
+    # Mirrors R Regression.R's tryCatch chain: NNS.dep error -> 0.1, copula
+    # error -> keep the NNS.dep component, NA -> 0.1.
+    try:
+        dep = nns_dep(x, y, asym=True)["Dependence"]
+    except (ValueError, FloatingPointError, ZeroDivisionError):
+        dep = 0.1
     try:
         scaled = np.column_stack(
             (