Benchmark times in README may underestimate "average case" speed of tested commands

[sashaweiss]

The current benchmarks in the README used hyperfine's --warmup argument, which helped tree, itree, and rg by warming up hardware caches. (This is stated in the README.) However, in the average case using itree, caches are unlikely to be full of filesystem data.

There is a noticeable speedup running itree multiple consecutive times, versus the first time. Benchmarking should also be done using the --prepare flag to clear caches, to more accurately estimate the "average case" time.