Unify transaction and witness merkle roots

[scravy]

Copied from PR#798 discussion

Also partially related to this PR. We are working on also introducing a merkle root for finalizer commits.
We will change their signatures so they will also have a witness program. If we proceed with the current approach, we will add +2 roots to the headers, one for finalizer commits and one for their witness.

@Gnappuraz had an idea that what if we don't introduce the new merkle root for finalizer commits but combine with existing one. Example:

    root
   /    \
txs      commits

So, in the header, we can still use one record, but in a p2p commits message we send besides commits also the hash of txs so we will be able to match with the root.

Then, if we pick this approach, a similar pattern can be applied to the witness root.

        witness root
       /            \
witness              commits witness

One edge-case is that we not always have commits. The idea is to duplicate then what you have on the left leaf to the right.

Now my question :)

If we proceed with that proposal (we don't find an issue with that). What if we instead of grouping witnesses and txs, we will group entities, e.g.:

    root
   /    \
txs      txs witnesses

        commits root
       /            \
commits              commits witnesses

With the second approach we have 2 possible advantages:

1. we don't need to duplicate left leaf in case commits are missing so the hash will always be of one particular entity, commits and not commits or txs
2. by looking at the header we can see if we have any finalizer commits in the block (because we have a separate root for them). With the first approach, you can't answer this question by looking at the header only.

[scravy]

I guess the proposed version has some merits:

              tx merkel                          commit merkel
              tree hash                            tree hash
             /        \                           /         \
            /          \                         /           \
    txid merkel      wtxid merkel   commits txid merkel   commits wtxid merkel
     tree hash        tree hash          tree hash              tree hash

But I think we need to evaluate which repercussions this has e.g. on light clients. Related to merkle trees are fast merkle trees which are (a) faster (b) more secure, which I would also like to see tackled for 1.0 (#18).

I could imagine, if we wanted to return to the size of 80 bytes in the header, to go even further:

                     hash of combines merkle treed
                      /                          \
                     /                            \
              tx merkel                          commit merkel
              tree hash                            tree hash
             /        \                           /         \
            /          \                         /           \
    txid merkel      wtxid merkel   commits txid merkel   commits wtxid merkel
     tree hash        tree hash          tree hash              tree hash

But also here I am wondering what this means for SPV/light clients.