[FEA] Compare performance of decompression engine (HW) versus decompression kernels (SW) on Blackwell

[GregoryKimball]

Is your feature request related to a problem? Please describe.
The NVIDIA Blackwell Architecture includes a "decompression engine". This hardware component complements the existing approach for decompression that relies on nvCOMP and specialized decompression kernels. Let's evaluate the decompression throughput for the HW and SW implementations across the cuIO benchmarks.

Blackwell’s Decompression Engine and ability to access massive amounts of memory in the NVIDIA Grace™ CPU over a high-speed link—900 gigabytes per second (GB/s) of bidirectional bandwidth—accelerate the full pipeline of database queries for the highest performance in data analytics and data science with support for the latest compression formats such as LZ4, Snappy, and Deflate.

The cuIO benchmarks for parquet files use the "cardinality" and "run_length" axes to control page encoding efficiency and compression ratio. It's likely that the page count, compression ratio and page size influence decompression throughput.