Differencing and averaging operators on C-grid

[vadmbertr]

With the current implementation, differencing and averaging operators are not always defined at the midpoint.

Todo:

• [x] remove the ability for the user to provide U and V grids, only the T grid. Means that the cyclogeostrophy function should take the SSH as argument (instead of Ug and Vg), compute the geostrophic currents, and then the cyclo ones (plus, feels better in terms of design IMO),
• [x] adopt NEMO indexing convention (https://www.nemo-ocean.eu/doc/img360.png),
• [x] re-write differencing and averaging operators (should apply to an axis, with a left or right padding depending on the interpolation point),
• [x] re-write the test case to make sure it fails in case of regression.