Always use a full tree

[bifurcation]

Based on #689

Closes #645

[TWal]

Using full tree + filtered direct path seems like a fine combo!

As you said in #645 it not only faster, but also makes things a lot easier to understand in various places of the protocol.

I also like a lot the two examples you added (the one for filtered direct path, and the one for parent-hash link), they have a nice pedagogical value!

[bifurcation]

Interim 2022-05-26:

• This will require some optimized implementations
• Have people done performance evaulations?
  • @bifurcation and @TWal did evaluation on parent hash, this should be easier
• How does this affect the secret tree?
  • Would change the structure to be a full tree
  • Always log(N) derivations to get to a leaf, instead of sometimes less
• TODO: Hold this for next week, @bifurcation and @TWal to do performance evaluation

[bifurcation]

Mailing list post with performance metrics:
https://mailarchive.ietf.org/arch/msg/mls/42tmYeZIJRAAZTxnSb6pzXKZGa0/

[bifurcation]

Interim 2022-06-02:

• @bifurcation presented performance data showing +/- ~20% on join
• @Bren2010 asked about removing leaf index from LeafNodeHashInput in order to allow for precomputing blank subtree hashes
  • Even with leaf indices, can pre-generate, they just have to be in-place
  • No action
• Clear to merge once editorial comments from @Bren2010 are addressed