[experimental: arrow ingestion] Undo sorting record hack

[thorfour]

FrostDB requires rows to be sorting for ingestion. To ingest arrow records we sorted them by writing them to a Parquet buffer, and then reading them back out into an arrow record. This reduced a lot of the CPU gains that we had obtained using Arrow record ingestion.

We should remove the conversion from series->parquet->arrow to just be series->arrow and sort that resulting record.

The arrow community seems to be working on a sorting implementation in the library: apache/arrow#34719