[FEA] Create a multi-threaded nvbenchmark for groupby_max

[GregoryKimball]

In the IO microbenchmarks for libcudf, we have parquet_reader_multithread, which is a tool for studying GPU saturation with higher host thread counts and one CUDA stream per thread (PTDS, per-thread-default-stream).

https://github.com/rapidsai/cudf/blob/branch-24.08/cpp/benchmarks/io/parquet/parquet_reader_multithread.cpp

Let's please add a similar benchmark for groupby_max that lets us run multiple groupby aggregations concurrently to reach higher GPU saturation.

https://github.com/rapidsai/cudf/blob/branch-24.08/cpp/benchmarks/groupby/group_max.cpp