Benchmarking io_uring(direct_io) configurations using fio

I am building a async IO library and wanted to figure out which features of io_uring are worth it when focusing on block device IO.

Hopefully this benchmark/approach can be useful for other people doing similar things.