Flaky test_a_uniform_change_shows_in_every_decile at half-penny shifts #2500

test_a_uniform_change_shows_in_every_decile (the aggregate-impact decile tests, from #2335) is flaky. It fails at seed=0, shift=0.005.

• Why: the comparison treats a change of half a penny or less as no change, so a uniform shift that small doesn't "show in every decile", as the test expects.
• Effect: the test can fail CI at random on any PR.
• Fix: have the strategy exclude shifts at or below the half-penny threshold, or assert on shifts above it only. Don't change the threshold.
• Found by: the hub's independent review of Compare two datasets on one model in the aggregate-impact tool #2425 (10/19).