existing column statistics can be missed

MakeTypedColumnStats returns min/max statistics guided by column order:

if TYPE_DEFINED_ORDER - min_value/max_value, and min/max otherwise.

correct behavior should be based on __isset flag.

**Reporter**: [Renat Valiullin](https://issues.apache.org/jira/secure/ViewProfile.jspa?name=rip.nsk@gmail.com) / @rip-nsk
#### PRs and other links:
- [GitHub Pull Request #36814](https://github.com/apache/arrow/pull/36814)

_{**Note**: *This issue was originally created as [PARQUET-2327](https://issues.apache.org/jira/browse/PARQUET-2327). Please see the [migration documentation](https://issues.apache.org/jira/browse/PARQUET-2502) for further details.*}

从 MakeTypedColumnStats 开始，查看 issue 中链接的 GitHub Pull Request #36814。检查 min/max 统计信息是如何选择的，并验证决定现有统计信息的是 __isset 标志，而不是列顺序；完成的标准是不会遗漏现有的列统计信息。

由索引模型根据 Issue 内容生成。