[Parquet] Automatically split batches to prevent overflow

### Describe the enhancement requested

#197 added an option to force columns to use the Large variants of String and Binary to handle columns with data larger than 2GB. We should take this one step further to automatically reduce the batch size of records to fit within int32 offsets if the column data is too large to fit within a single String/Binary column (data is more than 2GB).

### Component(s)

Parquet

Start by reading issue #197 and tracing the Parquet batch-writing path in the Go implementation. Done means automatically reducing record batch sizes when String or Binary column data would exceed int32 offsets, including coverage for data larger than 2GB.

Written by the indexing model from the issue text.