Design, usability and performance improvements for AVL tree

[ahmadsalim]

I have tried to adjust some things in the AVL tree implementation in order to make the design more coherent, and improve performance.

I have made the following changes:

• Separation of invariant from data structure, in order to get full erasure of invariant at runtime (to save performance from having unary nats stored everywhere).
• Removed Using variants (since they would have not been corrected if not used with the correct Ord instance anyway) and changed the Graph to not use them
• Removed the parameterisation over Ord, it seems that it had introduced unification issues when being used, and I thought we should only roll it back when it becomes easier to use (i.e., when equality over type class instances is easier to get).

[jfdm]

Ah cheers. I do not have time to look at this for a couple of days. Deadlines, Deadlines and Deadlines. But i will try to soon. As long as the tests complete, and my code that uses the graph still works, then it should be fine.

[ahmadsalim]

It is fine :). I did run make test and it said that all tests had passed and the Graph code should still work fine I believe.

[ahmadsalim]

Also, notice that at some point we can probably share the backend Tree implementation now and have a lot of common code between e.g. AVL Trees and RB Trees since the invariants are separate. It is only things that modify the tree that actually need to be written separately since they affect the invariants.

[jfdm]

about reducing the code duplication. That sounds fine, but it is something that I won't be looking at in much depth anyway, at least for the next few months.