Mixing length diagnostic for GGL

[ifenty]

Check to see if there is a mixed layer depth diagnostic for GGL corresponding with l_d(k=0).

From issue: #169 on MITgcm:

Based on the above, it seems that l_u and l_d vary as a function of depth. In the interior when stratification is constant, l_u = l_d. At the surface, l_u is presumably zero and l_d should be the distance from the surface to the base of the mixed layer.

I don't know how l_d and l_u are stored in the model but it seems that we would want to record l_d(k=0) for surface mixing depth.

[ifenty]

It turns out that the model doesn't implement GGL exactly as described in the original Gaspar et al. paper. One GGL implementation uses an approximation introduced in Blanke and Delecluse 1993, whereby the local vertical density gradient is used to calculate the mixing length based on local TKE. I suppose one could diagnose the ML depth using the Gaspar et al. crierion using vertical stratification and TKE. I think it's worth looking at. Another idea would be to find the depth level where the mixing length shrinks dramatically. Above that depth the mixing lengths would be long because stratification is weak and below that depth the strong stratification would suppress the mixing length. Of course that assumes something about the TKE distribution.