Negative graph parsing tests

The `KHR_interactivity` specification has many statements about early rejection of malformed graphs, i.e., parsing/linking errors rather than runtime `err` flows.

We need to have sample assets that intentionally violate those rules (ideally one by one) to enforce consistent engine behavior across the ecosystem.

Start by reviewing the KHR_interactivity specification's early-rejection rules and the repository's existing sample assets. Create intentionally malformed graph assets that isolate those parsing or linking violations, ideally one rule per asset. Done means the negative cases cover the identified rules and can enforce consistent engine behavior across the ecosystem.

Written by the indexing model from the issue text.