Missing Full fast-path for HAMT

[ekmett]

When the node in a HAMT is full, you can avoid all the popcounting overhead, which usually makes it worth special casing. Was sad when after 5 minutes of adding random nodes to finally get the root to be made of all 1s that it didn't switch from Many to another case, like we have in unordered-containers. ;)

[vaibhavsagar]

This is intentional! I wanted the implementation to be as simple as possible and I copied it straight from https://vaibhavsagar.com/blog/2018/07/29/hamts-from-scratch/. A production-ready implementation would support full nodes and collisions (as indicated in that blog post).

[vaibhavsagar]

@ekmett I updated it with support for Full in a6184f9, thanks for raising this!

[vaibhavsagar]

You can test this by using the numbers 0-9, 18-19, 22-23, and 26-27 as keys.