Define indefinite column integrals for arbitrary functions

[charleskawczynski]

ClimaAtmos currently uses

    prob = SciMLBase.ODEProblem(dp_dz, p_0, (FT(0), z_max))
    return SciMLBase.solve(
        prob,
        ODE.Tsit5(),
        reltol = 10eps(FT),
        abstol = 10eps(FT),
    )

to compute a vertical indefinite integral. We currently only need this for initialization, and CliMA/ClimaAtmos.jl#2183 implements this in a way that seems to work (MSE table changes are small). We could start by porting that over to ClimaCore and adding unit tests. Next (if we need) we could make this work on the GPU, which I think shouldn't be too difficult (it may work already as I've followed a similar pattern used for the other indefinite integral on fields).

[simonbyrne]

You added something in #1380: is there anything else we need?

[charleskawczynski]

You added something in #1380: is there anything else we need?

Here's a summary:

• Fix gpu column integrals #1380 (merged) added definite integrals (we still need to add indefinite integrals)
• Migrate CA column ops to ClimaCore #1445 adds indefinite integrals, but the interface is very different from what we need in ClimaAtmos
• Remove dependence on OrdinaryDiffEq ClimaAtmos.jl#2183 adds indefinite integrals the interface we need in ClimaAtmos (one input arg is an arbitrary function of z).

Both indefinite integrals need unit tests, it looks like @dennisYatunin pushed since the last bors try, so I just bors try'd again to see if he fixed the unit tests there. Maybe we can re-purpose them for the 3rd bullet implementation.