Rewrite in-memory Merkle tree

[pav-kv]

The code for internal/merkle/inmemory is old and needs a refresh. This change rewrites
the in-memory Merkle tree concisely, and updates all the call sites.

The previous implementation quoted C++ code, used a lot of non-Go constructs, and did
not have the benefit of reusing other Merkle primitives. All of these are good reasons for a
complete rewrite.

[pav-kv]

Review commit-by-commit to see the progression, and observe that the new code is equivalent.

[pav-kv]

Or review the whole thing, it can be clearer.

[pav-kv]

@AlCutter @mhutchinson I noticed that this code hasn't been moved to merkle repo. I think this implementation can a) be used for testing in many places, b) serve as a PoC implementation of a Merkle tree with compact and proof types (i.e. implementations with storage would be similar in principle, but have DB/files instead of the in-memory variable). A bit more polish (particularly in the area of unit-tests) for this type, and it can be placed in merkle.

[pav-kv]

ping

[mhutchinson]

Between being ill and oncall, I'm unlikely to get to this in the next few days. And I'm off next week. This is one for @AlCutter

[pav-kv]

@AlCutter I made a few more simplifications. It's strictly more convenient to review the latest state of the code. You may look at individual commits mainly to understand my "thinking process".