arrow

PARQUET-1857: [C++] Do not fail to read unencrypted files with over 32767 row groups. Change some DCHECKs causing segfaults to throw exceptions

While writing files with so many row groups is a bad idea, people will still do it and currently the C++ library will segfault in release builds when trying to read such a file. This removes those potential segfaults and enables reading the many-row-group files again. Files with encrypted row group metadata with that many row groups cannot be read because the Parquet metadata uses an int16 row group ordinal key.

Closes #7108 from wesm/PARQUET-1857

Authored-by: Wes McKinney <wesm+git@apache.org>
Signed-off-by: Wes McKinney <wesm+git@apache.org>