DataFrame API: to_parquet(partition_cols=) doesn't work as intended

[damccorm]

Currently we accept the partition_cols keyword argument, but it doesn't work as intended. It should partition by the specified columns and use dynamic destinations to write partitions to different files.

Context: https://lists.apache.org/thread.html/ra1e647440ffb43e922d9289cbe6f59e581c00055cf7f6a71b3fab205%40%3Cuser.beam.apache.org%3E

Imported from Jira BEAM-12201. Original Jira may contain additional context.
Reported by: bhulette.