Assert that we can compute the bounds for guards using rational numbers (#105139)
This makes sure that the bounds are always correct, as we're not losing
precision
Pull Request resolved: https://github.com/pytorch/pytorch/pull/105139
Approved by: https://github.com/ezyang
