Add test and improve inference of monotonic_check

[charleskawczynski]

This PR adds unit tests for monotonic_check, and inference tests (a step towards #1797). Previously, the closure z was getting captured inside any, and map was returning an object.

[Sbozzolo]

Thank you!

[charleskawczynski]

Unfortunately, I'm getting slightly different inference results locally compared to central for the cuda inference. I'll try to match the central numbers I'm going to pick the conservative numbers so that my local machine passes CI, too. Hopefully we can eventually drive this down to 0 on both. I'll rely on @test_opt for decision making on the CPU, since I think that is a better metric, anyway.