Add iteration and standard impls in once crate

[JoeyBF]

It was getting annoying that MultiIndexed didn't implement Debug, Clone, PartialEq, Eq, Hash, since any structs that contain one would not be able to derive them. The main bottleneck was being able to iterate over the entire tree.

I implemented iterators and updated the criterion bench to make sure the performance is close to optimal. I also implemented all the standard traits for MultiIndexed, KdTrie, Grove and TwoSidedGrove. I added proptests to check for correctness too.

[JoeyBF]

For future reference, I did try unwrapping the new methods in benchable_impls.rs, so that the impl for Benchable<K, T> would call the method for Benchable<K-1, T> in a flat_map. This makes me very confused as the unwrapped version is significantly faster for large K (~3x), but also significantly slower for K = 2 (~4x). I'll leave the implementation as-is because I feel it's the most representative for uses outside of this crate.

Profiling the bad case with perf shows that the code is now spending most of its time executing the AVX2 instruction vmovups outside of the iterator. Interestingly, this is also the instruction that gives the bottleneck for the new iterators. With the naive nested loops, the bottleneck is now just the atomic load in WriteOnce.