fix: optimize new set implementation and update docs

[austinabell]

There is no benefit to the cache in the set's case, as no APIs depend on it, and since no data loaded, the cache adds more overhead than benefit. This change also removes the Ord and Clone requirements on the set's keys.

Benchmarking these changes are ~15% decrease in gas for a small number of elements, and ~25% for larger amounts of modifications

[itegulov]

Benchmarking these changes are ~15% decrease in gas for a small number of elements, and ~25% for larger amounts of modifications

Interesting, would you mind sharing the benchmark source code?

In any case, LGTM and sounds likely that the caching overhead outweighs the benefits from what I remember from the benchmarks I did back in January.

[austinabell]

Interesting, would you mind sharing the benchmark source code?

Yeah, was hacked together quickly, and didn't update all the values or point to a git dep, but here it is: austinabell/storage_collections_bench@b3864e3