[C++][Parquet] A 175M DataFrame saved to parquet requires 1G of memory to be read

### Describe the bug, including details regarding any error messages, version, and platform.

```python
import memray
import pyarrow.parquet
import pandas as pd

myfile_path = "history_files.parquet"

with memray.Tracker("corr.bin", native_traces=True):
# myfile = pd.read_parquet(myfile_path)
myfile = pyarrow.parquet.read_table(myfile_path, use_threads=False)
```

This reports a peak of 1GiB. Is that expected?

[myfile.zip](https://github.com/apache/arrow/files/12882729/myfile.zip)
[flamegraph.zip](https://github.com/apache/arrow/files/12882735/flamegraph.zip)

### Component(s)

Python