Fix test flakiness

[MichaelChirico]

Closes #1350

Alternatively, we can use a set.seed(), but keeping this test random does make it slightly more interesting.

I also tried with n=150, it fails ~0.05% of the time. I didn't notice a speed difference on n=150/n=200, so I stuck with the safer n=200.

[MichaelChirico]

Another option is to deterministically ensure 1:15 are present, e.g. instead of sample.int(15, 100, TRUE), we do c(1:15, sample.int(15, 85, TRUE)).

[topepo]

"More interesting" 😄 I set the seed just to keep things more consistent.

Thanks for doing this.