Offspring FastDataset "leaking" add()s to parent

[Peeja]

It looks like a FastDataset derived from another one is meant to be completely independent (logically) of its parent, but still structurally share its parent's data as much as possible. If that's the intention, I believe I've found a bug.

Steps to Reproduce

1. Create a FastDataset containing quads about at least one subject.
2. Create a derivative FastDataset which includes at least one of those quads.
3. In the second dataset, add() a quad about the same subject.

Expected: The add()ed quad should exist in the second dataset, but not the first.
Actual: The add()ed quad appears in both datasets now.

See live example.

More info

• The original dataset will not pick up quads about any other subjects add()ed to the second dataset.
• The example uses match() to create the second dataset, but I've tested and seen the same behavior with minus(), and I assume it applies in other cases as well.