BUG: Fix handling of the dtype parameter to numpy.array_api.prod()

Original NumPy Commit: 19a398ae78c3f35ce3d29d87b35da059558d72b4

Author: asmeurer

array_api_strict/_statistical_functions.py

@@ -65,8 +65,8 @@ def prod(
     # Note: sum() and prod() always upcast float32 to float64 for dtype=None
     # We need to do so here before computing the product to avoid overflow
     if dtype is None and x.dtype == float32:
-        x = asarray(x, dtype=float64)
-    return Array._new(np.prod(x._array, axis=axis, keepdims=keepdims))
+        dtype = float64
+    return Array._new(np.prod(x._array, dtype=dtype, axis=axis, keepdims=keepdims))
 
 
 def std(