Remove mypy python_version pin to support newer numpy stubs

pyproject.toml

@@ -115,8 +115,10 @@ select = ["E", "F", "I", "B", "UP", "N", "RUF", "TID"]
 "scripts/inspect_r_api_update.py" = ["TID251"]
 "scripts/verify_release_provenance.py" = ["TID251"]
 
+# No python_version pin: mypy targets the interpreter it runs under. The CI
+# matrix runs mypy on 3.11-3.14, and pinning to 3.11 made mypy reject newer
+# numpy stubs that use PEP 695 `type` statements (valid only on 3.12+).
 [tool.mypy]
-python_version = "3.11"
 strict = true
 files = ["src/nns", "tests"]
 mypy_path = ["tests"]

src/nns/central_tendencies.py

@@ -135,10 +135,12 @@ def _discrete_mode(values: NDArray[np.float64], multi: bool) -> float | NDArray[
     integerized = _nearest_int_half_up_array(values)
     modes, counts = np.unique(integerized, return_counts=True)
     tied_modes = modes[counts == int(np.max(counts))]
-    tied_modes = np.sort(tied_modes.astype(np.float64))
+    # Annotate the sorted result so the multi return is typed rather than Any
+    # under the more precise numpy stubs that mypy sees when targeting 3.12+.
+    sorted_modes: NDArray[np.float64] = np.sort(tied_modes.astype(np.float64))
     if multi:
-        return tied_modes
-    return float(np.mean(tied_modes))
+        return sorted_modes
+    return float(np.mean(sorted_modes))
 
 
 def _continuous_mode(values: NDArray[np.float64], multi: bool) -> float | NDArray[np.float64]:

src/nns/norm.py

@@ -22,7 +22,12 @@ def nns_norm(x: NDArray[np.float64], linear: bool = False) -> NDArray[np.float64
         scale_factor = _scale_factor(values)
         scales = np.mean(ratio_grid * scale_factor, axis=0)
 
-    return cast(NDArray[np.float64], values * scales[np.newaxis, :])
+    # Annotated assignment instead of cast(): under mypy targeting 3.12+ the
+    # numpy stubs type this product precisely, making an explicit cast redundant
+    # (warn_redundant_casts), while on 3.11 the stubs return Any and the
+    # annotation still narrows it without a no-any-return error.
+    scaled: NDArray[np.float64] = values * scales[np.newaxis, :]
+    return scaled
 
 
 def _scale_factor(values: NDArray[np.float64]) -> NDArray[np.float64]: