Right now, we read bytes from DistSQL and buffer bytes in the parquet library before writing them. We don't report the fact that we are buffering these bytes. EXPORT INTO PARQUET should report that these allocs are being held to memory monitors.

Jira issue: CRDB-28470