[SPARK-48310][PYTHON][CONNECT] Cached properties must return copies

### What changes were proposed in this pull request?
When a consumer modifies the result values of a cached property it will modify the value of the cached property.

Before:
```python
df_columns = df.columns
for col in ['id', 'name']:
  df_columns.remove(col)
assert len(df_columns) == df.columns
```

But this is wrong and this patch fixes it to

```python
df_columns = df.columns
for col in ['id', 'name']:
  df_columns.remove(col)
assert len(df_columns) != df.columns
```

### Why are the changes needed?
Correctness of the API

### Does this PR introduce _any_ user-facing change?
No, this makes the code consistent with Spark classic.

### How was this patch tested?
UT

### Was this patch authored or co-authored using generative AI tooling?
No

Closes #46621 from grundprinzip/grundprinzip/SPARK-48310.

Authored-by: Martin Grund <[email protected]>
Signed-off-by: Hyukjin Kwon <[email protected]>

python/pyspark/sql/connect/dataframe.py

@@ -43,6 +43,7 @@
     Type,
 )
 
+import copy
 import sys
 import random
 import pyarrow as pa
@@ -1787,7 +1788,7 @@ def schema(self) -> StructType:
         if self._cached_schema is None:
             query = self._plan.to_proto(self._session.client)
             self._cached_schema = self._session.client.schema(query)
-        return self._cached_schema
+        return copy.deepcopy(self._cached_schema)
 
     def isLocal(self) -> bool:
         query = self._plan.to_proto(self._session.client)

python/pyspark/sql/tests/connect/test_parity_dataframe.py

@@ -19,13 +19,37 @@
 
 from pyspark.sql.tests.test_dataframe import DataFrameTestsMixin
 from pyspark.testing.connectutils import ReusedConnectTestCase
+from pyspark.sql.types import StructType, StructField, IntegerType, StringType
 
 
 class DataFrameParityTests(DataFrameTestsMixin, ReusedConnectTestCase):
     def test_help_command(self):
         df = self.spark.createDataFrame(data=[{"foo": "bar"}, {"foo": "baz"}])
         super().check_help_command(df)
 
+    def test_cached_property_is_copied(self):
+        schema = StructType(
+            [
+                StructField("id", IntegerType(), True),
+                StructField("name", StringType(), True),
+                StructField("age", IntegerType(), True),
+                StructField("city", StringType(), True),
+            ]
+        )
+        # Create some dummy data
+        data = [
+            (1, "Alice", 30, "New York"),
+            (2, "Bob", 25, "San Francisco"),
+            (3, "Cathy", 29, "Los Angeles"),
+            (4, "David", 35, "Chicago"),
+        ]
+        df = self.spark.createDataFrame(data, schema)
+        df_columns = df.columns
+        assert len(df.columns) == 4
+        for col in ["id", "name"]:
+            df_columns.remove(col)
+        assert len(df.columns) == 4
+
     @unittest.skip("Spark Connect does not support RDD but the tests depend on them.")
     def test_toDF_with_schema_string(self):
         super().test_toDF_with_schema_string()