Nestable exceptions shouldn't be raised if the experiment isn't being "run"

Based on some discussions that happened around https://gitlab.com/gitlab-org/gitlab/-/issues/344598#note_755329443, I think what's considered running an experiment is too loose here for nesting logic, and we shouldn't raise the exception under certain circumstances.

We shouldn't raise the exception if #should_track? returns false, which means that we're effectively not running, tracking, and therefore don't care about the nesting logic.