[FEA] Follow-up items for benchmarks PR

[cjnolet]

The initial version of our ANN benchmarks leaves some things to be desired. I'm capturing them here as a tracker so we can quickly follow-up on them (likely in 23.06 when things calm down a bit):

• Convert pearl script to Python
• Use RMM (and corresponding RAFT utilities) throughout benchmarking code
• Port over Python client scripts (like run.py, install.py, and plot.py from big-ann-benchmarks.) [DOC] Simplify bench-ann scripts and documentation #1633
• Use file/directory management functions in std library and remove mannual file management functions
• Replace logging statements w/ RAFT logger
• Check fp16 support (especially for FAISS)