NBP constraint crashes model in some circumstances

[bpbond]

@jk-brown and @kendalynnm have found a case where supplying the model with an NBP constraint crashes it. 💥

Our automated tests currently check that things are OK with a 21st-century constraint...but giving a 100-year 20th century constraint causes a crash.

[bpbond]

The proximal cause of the crash was a fluxpool negative-value error. This is easy to fix/work around...

But the deeper issue is that there's some issue with the NBP constraint logic, and the model is having to work increasingly hard to meet it the constraint (eventually transferring tens of Pg C to the deep ocean before crashing with a mass balance error in the middle of the 21st century). 😕

[bpbond]

Stopping point for now

• Comparing constrained and free runs, atmosphere pool visibly diverging by 1763
• Do we want to remove NBP constraint consideration from calc_derivs? Think about this