Consider using const generics for HAMT degree

[djudd]

This would be a breaking change and so I assume would require a major version bump, and probably doesn't warrant that on its own, but may be worth considering if & when there is a new major version. I'd expect some degree of optimization possibility - for example, IIUC it would allow https://github.com/orium/rpds/pull/108 to be replaced with just using an array - and I would be surprised if anyone is relying on creating HAMTs of different degree at the same callsite in practice.