Add convergence test for indefinite integrals

[charleskawczynski]

We should add spatial convergence tests to our indefinite integrals. Cross ref discussion: #1504 (comment). Doing this requires collecting the errors and, at the moment this involves 2 loops already, maintaining a 3-tuple dictionary seems like a bit overkill (we need to collect the errors to compute the convergence rate). Maybe there's a simple way we can do this, but it requires a bit of thought. I've already (offline) confirmed that the operator is 2nd order accurate. Also, the existing tests in ClimaAtmos fail with small MSE error changes, so I think this is closely approximating SciMLBase.solve.