test_log_margins test failure

[Kojoley]

It was broken by #7144. The tests passed in the PR because at the point when they were made there was no test_log_margins test (it was restored by #7435).

[efiring]

I think the solution might be to add an atol=1e-16 argument to the assertion. It looks like the problem is that it is using an atol of 0, which allows for no floating point slop in a comparison to 0.
Or, change each [1, 10] to [10, 100], so that the test values would be (1, 2) instead of (0, 1).
(I haven't tried either of these option.)