CI should have a job that does relative (percentage deltas) and absolute size regressions of the demos.

CI should fail if the PR introduces a regression beyond the agreed upon thresholds.

The analysis should be done on the final demo sizes, and the reduces test_binary_size target.