Qwen Image 1140x1472 (and mirror) and SD 1.5 512x682 were not divisible
by 8, so the generated latent did not match the reported width/height;
removed the stray Qwen pair (official list has 1104x1472) and corrected
682 to 680. The preset branch now runs _validate_dimensions before the
multiplier, so API-built workflows cannot pair a model with another
model's incompatible preset. Regenerated the hand-copied JS table from
nodes.py (it had drifted by 14 entries) and added tests that fail on
future drift or non-conforming presets. Bump to 2.1.6.