The path returned by `EmptySubtreeRoots` starts at the root, and goes to
the leaf. The MerkleStore constructor assumed the other direction, so
the parent/child hashes were reversed.

This fixes the bug and adds a test.

And unit tests for each constructor type.

Returning tuples is often confusing as they don't convey meaning and it
should be used only when there is no possible ambiguity.

For `MerkleStore`, we had a couple of tuples being returned, and reading
the implementation was required in order to distinguish if they were
leaf values or computed roots.

This commit introduces two containers that will self-document these
returns: `RootPath` and `ValuePath`. It will also update `set_node` to
return both the new root & the new path, so we can prevent duplicated
traversals downstream when updating a node (one to update, the second to
fetch the new path/root).

The store builds the path from root to leaf, this updates the code to
return a path from leaf to root, as it is done by the other structures.

This also added custom error for missing root.

Prior to this commit, the MerkleStore panicked under certain bounds. It
will prevent such panics by using checked operations.

ilog2, for instance, will panic when the operand is zero. However, there
is a documentation rule enforcing the merkle tree to be size at least 2.
If this rule is checked, then the panic is impossible.

This commit introduce common derive proc macros to MerkleStore. These
are required downstream as the in-memory storage can be cloned.

It also introduces constructors common to the other types of the crate
that will help to build a merkle store, using a build pattern.