CICE interface modification for initial CICE output

[dupontf]

Context

the NEMO-CICE interface (tag 5516) is missing a correct initialization for the CICE initial output. Some impact as well on the initial ice temperature profile. Some cleanup and consistency between CICE4 and CICE5

Analysis

CICE does an initialization based on default fields (exception: CICE5 is passed the correct sst) which could be based on NEMO fields. This can impact the initial ice temperature profile and the initial CICE output.
CICE4.0 takes default values for sss,sst,potT,Tair fields.
CICE5 takes values for sss,potT,Tair fields (sst is passed before the start of CICE init phase)

Fix

add a second init and CICE output after the correct sss,sst,Tf,potT,Tair fields are passed. (See attached file)
some other minor cosmetic changes such as:
1) the array ztmp is useless between 396 and 416
2) mass ice embedding has lines redundant with dom_vvl_init. Since those could useful as well in case of hot initialization, I suggest moving them in a distinct routine in domvvl.F90 so that they can be called anytime needed.
The only difference is
dom_vll_init has
fse3t_a(:,:,jpk) = e3t_0(:,:,jpk)
whereas sbcice_cice has
fse3t_a(:,:,:) = fse3t_b(:,:,:)
the latter being more consistent to my opinion and has no impact when ssh=0 (which is assumed when calling dom_vvl_init)

additional comment: could be merged with tiket 1428 of Tim Graham, since I also believe that a consistent calculation of the freezing temperature is needed in NEMO and CICE.