Use size classes instead of doubling by powers of 2

With powers of 2, there's a large average waste for allocations which is around 25% (assuming a random distribution of sizes between 1 byte and the small size limit).

Would there be any interest in adding size classes similar to for example Go as listed in https://github.com/golang/go/blob/master/src/runtime/sizeclasses.go?

We've been playing with that locally with code forked from this library, but maybe there's interest in having those changes here as well?