This removes the `mark` function from the `tree` interface, in favor
of always appending nodes as marked, as that's what's needed in
practice, rather than the ability to mutably mark the latest position at
arbitrary states of the tree.
This removal does mean that a few patterns of interleaved mark and
checkpoint operations are no longer being tested in the shared test
suite; however, such interleaving of operations is not something that
we should need to support anyway.
This change exposed an inconsistency in how `BridgeTree` and
`CompleteTree` were tracking `Mark` operations, which has now also been
fixed.
The behavior of the "unmark" operation has been simplified, such that it
now ensures that a marked node will be made eligible for garbage
collection when the checkpoint for the tree state in which the mark was
forgotten rolls off the back of the checkpoint queue.
Kris Nuttycombe