Merging two valid parquet files produces a corrupted result file in 1.12.1

This ticket relates to PARQUET-2027. In the previous ticket for two parquet files produced by 1.11.x merging was failing in 1.12.0. For 1.12.1 merging was fixed, i. e. it doesn't fail. But in the same time it results with a corrupted output file. The error:
```java

Dictionary page must be before data page.
```
is thrown when one tries to read it. It comes from this .

I attached two example input files and the outcome of merging.

**Reporter**: [Matthew M](https://issues.apache.org/jira/secure/ViewProfile.jspa?name=eltherion)
#### Original Issue Attachments:
- [input_file1.parquet](https://issues.apache.org/jira/secure/attachment/13034644/input_file1.parquet)
- [input_file2.parquet](https://issues.apache.org/jira/secure/attachment/13034643/input_file2.parquet)
- [output_file.parquet](https://issues.apache.org/jira/secure/attachment/13034642/output_file.parquet)

_{**Note**: *This issue was originally created as [PARQUET-2100](https://issues.apache.org/jira/browse/PARQUET-2100). Please see the [migration documentation](https://issues.apache.org/jira/browse/PARQUET-2502) for further details.*}

Riproduci il merge con input_file1.parquet e input_file2.parquet usando 1.12.1, quindi prova a leggere output_file.parquet per confermare l'errore della pagina del dizionario. Confronta il fallimento con PARQUET-2027 e ispeziona la riga 712 indicata di parquet-cpp record_reader.cc. Il lavoro è completato quando l'output sottoposto a merge è leggibile senza la corruzione segnalata.

Scritto dal modello di indicizzazione a partire dal testo della issue.