continuation of#3199

Pretty important PR since sub-rowgroup loading lets us:

• load Pandas datasets that are >300MB, or from parquet files file PyArrow/Spark if row groups are >300MB
• increase the row group size indatasets for better Xet deduplication

TODO:

• create page index when required (row groups > 300MB)
  • it's a richer parquet metadata file than the one from from the "config-parquet-metadata" job and it replaces the existing metadata file
• load rows using page index when required

It works using page pruning fromarrow-rs