RapidsMPF runtime OOMs in some PDS-H queries

[rjzamora]

Follow up to #618

Several TPC-H queries now hit OOM or segmentation-fault errors with the following settings on PDX (1xH100):

• --executor="streaming"
• runtime="rapidsmpf"
• path="/data/tpch-rs/scale-1000"
• --suffix=""
• --stream-policy="pool"
• --n-workers 1
• -no-print-results
• --no-summarize
• --iterations=2
• --rmm-async
• --blocksize 2_000_000_000

Failures (example output):

• q1
• q4
• q13
• q17
• q18

[rjzamora]

cc @TomAugspurger (feel free to modify or add to this issue)

[rjzamora]

rapidsai/cudf#20615 will bump q17 off the "problem" lest.

[rjzamora]

rapidsai/cudf#20567 will bump q1 off the "problem" list.