This commit introduces a method that corrects the remapped velocity so
that it conserves KE. The correction is activated by setting
`REMAP_VEL_CONSERVE_KE = True`
The commit also introduces two new diagnostics:
`ale_u2` and `ale_v2`
These track the change in depth-integrated KE of the u and v components
of velocity before the correction is applied. They can be used even
if the remapping correction is not turned on.

This commit does two main things.
- Limit the magnitude of the multiplicative correction applied to
the baroclinic velocity to +25%. This prevents rare occasions where
the correction creates very large baroclinic velocities.
- Move the diagnostic of KE loss/gain from before the correction
to after the correction. Without the limit (above) the correction
is exact to machine precision, so there was no point in computing
it after the correction. With the new limit it makes sense to
compute the diagnostic after the correction.

This commit addresses @Hallberg-NOAA's comments on
[the PR](NCAR#277). Computations of
`ale_u2` and `ale_v2` are updated to work correctly in both
Boussinesq and non-Boussinesq modes.