[SPARK-40590][TEST] Fix ps.read_parquet when pandas_metadata is True

### What changes were proposed in this pull request?

This PR proposes to fix the `ps.read_parquet` test since the `pd.to_parquet` is broken when the index is `MultiIndex` from pandas 1.5.0.

We leverage the `pd.to_parquet` in the test, so the test failed with pandas 1.5.0 as below (`MultiIndex` is not respected):
```python
DataFrame shape mismatch
[left]:  (20, 5)
[right]: (20, 4)

Left:
    i32  i64     f  bhello     index
0     0    0   0.0      yo  0.617492
1     1    1   1.0  people  0.823826
2     2    2   2.0  people  0.443275
3     0    3   3.0   hello  0.639776
4     1    4   4.0      yo  0.393410
5     2    0   5.0      yo  0.898860
6     0    1   6.0  people  0.725236
7     1    2   7.0      yo  0.933009
8     2    3   8.0      yo  0.663381
9     0    4   9.0   hello  0.471077
10    1    0  10.0   hello  0.562182
11    2    1  11.0  people  0.734902
12    0    2  12.0      yo  0.956519
13    1    3  13.0   hello  0.860517
14    2    4  14.0  people  0.012749
15    0    0  15.0  people  0.561815
16    1    1  16.0  people  0.389130
17    2    2  17.0   hello  0.930301
18    0    3  18.0   hello  0.835025
19    1    4  19.0      yo  0.212191
i32         int32
i64         int64
f         float64
bhello     object
index     float64
dtype: object

Right:
             i32  i64     f  bhello
   index
0  0.617492    0    0   0.0      yo
1  0.823826    1    1   1.0  people
2  0.443275    2    2   2.0  people
3  0.639776    0    3   3.0   hello
4  0.393410    1    4   4.0      yo
5  0.898860    2    0   5.0      yo
6  0.725236    0    1   6.0  people
7  0.933009    1    2   7.0      yo
8  0.663381    2    3   8.0      yo
9  0.471077    0    4   9.0   hello
10 0.562182    1    0  10.0   hello
11 0.734902    2    1  11.0  people
12 0.956519    0    2  12.0      yo
13 0.860517    1    3  13.0   hello
14 0.012749    2    4  14.0  people
15 0.561815    0    0  15.0  people
16 0.389130    1    1  16.0  people
17 0.930301    2    2  17.0   hello
18 0.835025    0    3  18.0   hello
19 0.212191    1    4  19.0      yo
i32         int32
i64         int64
f         float64
bhello     object
dtype: object
```

### Why are the changes needed?

We should make the all test passing with pandas 1.5.0.

### Does this PR introduce _any_ user-facing change?

No.

### How was this patch tested?

Manually test with pandas 1.5.0.

Closes #38055 from itholic/SPARK-40590.

Lead-authored-by: itholic <haejoon.lee@databricks.com>
Co-authored-by: Haejoon Lee <44108233+itholic@users.noreply.github.com>
Signed-off-by: Hyukjin Kwon <gurwls223@apache.org>

Author: itholic

python/pyspark/pandas/tests/test_dataframe_spark_io.py

@@ -18,6 +18,7 @@
 import unittest
 import glob
 import os
+from distutils.version import LooseVersion
 
 import numpy as np
 import pandas as pd
@@ -96,11 +97,18 @@ def test_parquet_read_with_pandas_metadata(self):
             self.assert_eq(ps.read_parquet(path2, pandas_metadata=True), expected2)
 
             expected3 = expected2.set_index("index", append=True)
+            # There is a bug in `to_parquet` from pandas 1.5.0 when writing MultiIndex.
+            # See https://github.com/pandas-dev/pandas/issues/48848 for the reported issue.
+            if LooseVersion(pd.__version__) == LooseVersion("1.5.0"):
+                expected_psdf = ps.read_parquet(path2, pandas_metadata=True).set_index(
+                    "index", append=True
+                )
+            else:
+                path3 = "{}/file3.parquet".format(tmp)
+                expected3.to_parquet(path3)
+                expected_psdf = ps.read_parquet(path3, pandas_metadata=True)
 
-            path3 = "{}/file3.parquet".format(tmp)
-            expected3.to_parquet(path3)
-
-            self.assert_eq(ps.read_parquet(path3, pandas_metadata=True), expected3)
+            self.assert_eq(expected_psdf, expected3)
 
     def test_parquet_write(self):
         with self.temp_dir() as tmp: