Update dataset interpolation procedure

[gmaze]

The ADMT has recently endorsed a clean vertical interpolation procedure for T/S profiles.

The procedure is described in the EasyOneArgo data product specification and the interpolation method is based on "multiple Piecewise Cubic Hermite Interpolating Polynomials" as introduced by Barker and McDougall method, which is effective in avoiding the production of spurious water masses from interpolation of T and S separately.

The goal of this PR is to implement this new interpolation procedure in Argopy.

The current interpolation method is described here:
https://argopy.readthedocs.io/en/latest/user-guide/working-with-argo-data/data_manipulation.html#pressure-levels-interpolation

The current API is:

press_levels = [0,10,20,30,40,50]
ds_interp = ds.argo.interp_std_levels(press_levels)

which relies on the argopy.utils.linear_interpolation_remap function.

The new API would use the xarray dataset extension design, and provides a new option to select the interpolation method:

ds_interp = ds.argo.interp.spl(press_levels)
ds_interp = ds.argo.interp.spl(press_levels, method='pchip') # default, new EasyOneArgo method
ds_interp = ds.argo.interp.spl(press_levels, method='linear') # Method used in < v1.4.0 (i.e. linear_interpolation_remap)

To complete the EasyOneArgo integration, we could also add the possibility to provide a string instead of an array with pressure levels:

ds.argo.interp.spl('EasyOneArgoLite') # Use pre-defined pressure levels from EasyOneArgo

Rq: Update to BGC variables interpolation will be addressed in another PR

PR to-do list

• Implement new interpolation procedure for core T/S
• Validate methods against EasyOneArgo product
• Check how to interpolate BGC variables with legacy or new functions (to be done in another PR)
• Document deprecation of the legacy API
• Add CI tests
• Update documentation