I don't really like passing the block, but this seems easiest for now

turns out that the activerecord object will always have the same id as
the row from which it was constructed

Remove duplicate removal code because we avoid adding duplicates as we
walk the tree the first time

We don't need to poke at the macro to figure out if we should look for a
cached record or not

For now, we'll set the tables on the nodes manually.

flatten and flatten! methods for ActiveSupport::Duration

No need to require rbconfig as we are using Gem.ruby

we don't need to walk back up the node's parents when we have the parent
on the stack

Now that we merge trees from the top down, we don't need to search
through the whole tree for particular nodes, just walk it and merge.

Merge JoinDependency objects as outer joins