Investigate feruca's performance claim

[hsivonen]

feruca claims:

feruca performs on the order of 2–4x faster than the icu4x collator—while having a much smaller feature set.

The claim has been updated after ICU4X 2.0.

At minimum, plug feruca into icu_collator's root collation benches, see what performance difference they show, and investigate the cause.

[hsivonen]

https://github.com/theodore-s-beers/feruca-benchmarks appears to be what the performance claim is based on. The workload seems even less realistic than our synthetic bench workloads.