[BUG] Fix .str.length() on Unicode strings (#2579)

Previously, the `.str.length()` method would count the number of bytes
in the UTF-8 string. This is inconsistent with Python's `len()` and
pandas' `str.len()` which count Unicode codepoints. For instance, on the
string "😉test", the number of bytes is 8, whereas the number of
codepoints is 5. This PR makes Daft consistent with that behavior.

There doesn't seem to be a way now to reproduce the original behavior;
maybe we should add a `.byte_length()` method for that.

src/daft-core/src/array/ops/utf8.rs

@@ -610,7 +610,7 @@ impl Utf8Array {
             .iter()
             .map(|val| {
                 let v = val?;
-                Some(v.len() as u64)
+                Some(v.chars().count() as u64)
             })
             .collect::<arrow2::array::UInt64Array>()
             .with_validity(self_arrow.validity().cloned());

tests/series/test_utf8_ops.py

@@ -265,6 +265,12 @@ def test_series_utf8_length_all_null() -> None:
     assert result.to_pylist() == [None, None, None]
 
 
+def test_series_utf8_length_unicode() -> None:
+    s = Series.from_arrow(pa.array(["😉test", "hey̆"]))
+    result = s.str.length()
+    assert result.to_pylist() == [5, 4]
+
+
 @pytest.mark.parametrize(
     ["data", "expected"],
     [

tests/table/utf8/test_length.py

@@ -5,6 +5,6 @@
 
 
 def test_utf8_length():
-    table = MicroPartition.from_pydict({"col": ["foo", None, "barbaz", "quux"]})
+    table = MicroPartition.from_pydict({"col": ["foo", None, "barbaz", "quux", "😉test", ""]})
     result = table.eval_expression_list([col("col").str.length()])
-    assert result.to_pydict() == {"col": [3, None, 6, 4]}
+    assert result.to_pydict() == {"col": [3, None, 6, 4, 5, 0]}