make gp tests almost equal
The CI has started failing ~10% of the time due to the recently added GP likelihood tests, e.g., here. This MR replaces the current test for equality with an approximate test, hopefully, this will lead to fewer random failures.