Redesign and/or remove support for Numerical coalescence style

[edejong-caltech]

Because quadgk is causing so many issues with optimization tests (and for future GPU support), we need to either rewrite our numerical coalescence style to use a different integration package; OR remove support altogether.

[trontrytel]

Let me know what you ended up doing. We do use quadgk in CM.jl for one of the P3 scheme integrals...

@charleskawczynski - would it make sense to take the quadratures and quadloop out of the ClimaAtmos depths and make it into something that different repos can use for numerical integrals?

[charleskawczynski]

Yeah, I think that would be useful. What are the errors we're seeing with quadgk?