[Kernel][Defaults] Support reading parquet files with legacy 3-level repeated types

[vkorukanti]

Description

When legacy mode is enabled in Spark, array physical types are stored slightly different from the standard format.

Standard mode (default):

optional group readerFeatures (LIST) {
  repeated group list {
    optional binary element (STRING);
  }
}

When write legacy mode is enabled (spark.sql.parquet.writeLegacyFormat = true):

optional group readerFeatures (LIST) {
  repeated group bag {
    optional binary array (STRING);
  }
}

TODO: We need to handle the 2-level lists. Will post a separate PR. The challenge is with generating or finding the Parquet files with 2-level lists.

How was this patch tested?

Added tests

Fixes #3082