The video size-target and zip Paranoia tests were not covered elsewhere:
the 10/20 MiB targets set the re-encode threshold, and only the Paranoia
test round-trips lineup and deleted Paranoias. Both are back. The dense
cone query keeps its ray and edge counters without the benchmark loop.

The fixture round-trip now reimports the exported .ica itself and
compares image attachments, so an exporter that drops images fails.
Imports and exports write their images inside the test's temp library,
and the Hive group deletes its temp library instead of leaking one per
run.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>