ParquetReader should check creator version before applying row group statistics

**Describe the bug**
as described in https://issues.apache.org/jira/browse/PARQUET-251, parquet files created by parquet-mr version <= 1.6.0 might contain incorrect row-group statistics.

newer parquet-mr version has added an extra logic to check parquet file creator, if it is created by buggy parquet-mr version. the statistics will be completely dropped. (the code is here: https://github.com/apache/parquet-java/blob/68e0eae9fb375d9b586ab3899828968f2a6d711e/parquet-column/src/main/java/org/apache/parquet/CorruptStatistics.java). we should keep the same logic with this implementation.

**To Reproduce**
query with buggy parquet file created by parquet-mr 1.6.0. the row group statisitcs is corrupted and results incorrect pruning and outpus.

**Expected behavior**
check parquet file creator, if created by buggy parquet-mr version, don't use its row group statistics.

**Additional context**
trino has fixed this bug: https://github.com/trinodb/trino/issues/1798

Start by locating the Rust ParquetReader row-group statistics pruning path and compare its behavior with parquet-java's linked CorruptStatistics.java. Reproduce the issue using a file created by parquet-mr 1.6.0, then verify that statistics from affected creators are ignored and incorrect pruning no longer occurs.

Written by the indexing model from the issue text.