fix(treebuild): use Unicode compatibility API

Use the existing kind-aware constructor wrapper for borrowed text spans.
PyPy requires its widening path to preserve surrogate code points; CPython
expands it to the same constructor used by the measured optimization.

Cover lone surrogates and surrogate pairs through tree-builder callbacks.

Author: gaborbernat

src/turbohtml/_c/dom/treebuild.c

@@ -81,7 +81,7 @@ static PyObject *node_str(th_tree *tree, th_node *node) {
     if (node->type != TH_NODE_DOCTYPE) {
         if (text_is_span(node)) {
             const char *data = (const char *)tree->data + text_span_offset(node) * tree->kind;
-            return PyUnicode_FromKindAndData(tree->kind, data, node->text_len);
+            return th_str_from_kind(tree->kind, data, node->text_len);
         }
         return PyUnicode_FromKindAndData(PyUnicode_4BYTE_KIND, node->text, node->text_len);
     }

tests/dom/test_treebuild.py

@@ -105,8 +105,17 @@ def test_element_carries_html_namespace_and_attribute_pairs() -> None:
 
 @pytest.mark.parametrize(
     "data",
-    ["hello", "café", "水", "😀", "before\nnext", "a&b"],
-    ids=["ascii", "latin1", "bmp", "astral", "newline", "entity"],
+    [
+        pytest.param("hello", id="ascii"),
+        pytest.param("café", id="latin1"),
+        pytest.param("水", id="bmp"),
+        pytest.param("😀", id="astral"),
+        pytest.param("before\nnext", id="newline"),
+        pytest.param("a&b", id="entity"),
+        pytest.param("\ud800", id="high-surrogate"),
+        pytest.param("\udfff", id="low-surrogate"),
+        pytest.param("\ud83d\ude00", id="surrogate-pair"),
+    ],
 )
 def test_text_node_payload(data: str) -> None:
     body = build(f"<p>{data.replace('&', '&amp;')}</p>").children[0].children[1]