Implement and test vertical mass borrowing limiters

[charleskawczynski]

This is a work in progress. This PR implements vertical mass borrowing limiters.

Closes #2083.

Todo:

• How should q_min be computed? I think q_min is actually just prescribed.
• Is pdel the correct interpretation? I think that this is just dz
• Add meaningful tests. Our tests currently operate on a specific total humidity. Should we also test it on ρq?
• Add GPU implementation
• Improve performance?

[charleskawczynski]

This does seem to work in that it removes negative tracer values, and seems to preserve mass (both of which are unit-tested cc @tapios):

On a single column:

    @test 0 ≤ minimum(q)
    @test isapprox(sum(q), sum_q_init; atol = 0.00000000001)
    @test isapprox(sum(q), sum_q_init; rtol = 0.01)

On a full cubed sphere:

    @test 0 ≤ minimum(q)
    @test isapprox(sum(q), sum_q_init; atol = 0.013)
    @test isapprox(sum(q), sum_q_init; rtol = 0.01)

I think rtol = 0.01 should be sufficient on a sphere, right?

But, the transport/wavespeed doesn't look great:

I guess that's a bit orthogonal, though?

[charleskawczynski]

pdel is a hydrostatic approximation of rho*dz, so we need to update this.

Also, the mass conservation test at the end should be rho*q instead.

[tapios]

Mass integral from rho dz, rather than dp

[charleskawczynski]

And let's visualize how this looks on the sphere.