Inside of the b+tree cow, we use entry dup functions. this may waste a lot of memory. As well, we assert that changes to the tree aren't made: Either an delete / insert happens which would cause a new value associated with key (in a new node), which would cause old values to be freed.

A better way, would be to create an internal reference counting box, that ref counts the values and keys in use. This way when they drop to 0, we implicitly free them. It allows external users to gain memory leak safety, and allows us to clamp our memory usage.

As part of this, an "update" or "replace" function should be added that implies delete / insert, but where the value associated to a key has it's ref count dropped, then replaced with a new box.