Performance improvement in mapping function

[delphinedobler]

This pull request contains three steps of performance improvements:
3.1 - Performance: save boxes_id from get_region_ow and recompute the index to provide as input to retr_region_ow
3.2 - Performance: Inside retr_region_ow, save loaded data, and at each following profile, reload only new boxes and delete unused.
3.3 - Performance: Inside map_data_grid, comment calculation of vdataerror as this is never used afterwards.

Improvement and integrity has been tested using wmo 3902007 until cycle 89, with the default ow_config.txt from the master branch: (300 max casts, MAP_USE_PV=0, MAP_USE_SAF=0, MAPSCALE_LONGITUDE_LARGE=6, MAPSCALE_LONGITUDE_SMALL=3, MAPSCALE_LATITUDE_LARGE=3, MAPSCALE_LATITUDE_SMALL=1, MAPSCALE_PHI_LARGE=0.5, MAPSCALE_PHI_SMALL=0.1, MAPSCALE_AGE_LARGE=20, MAPSCALE_AGE_SMALL=10, MAP_P_EXCLUDE=200, MAP_P_DELTA=250) and latest climatology to date (CTD (2024v01) and Argo (2023v03)).

In the testing case, it was observed an average 9.7 wmo boxes used for each profile (min 8, max 12).

The following execution times were observed:
initial execution : 18.63 minutes
3.1 change : 12.65 minutes
3.2 change : 8.77 minutes
3.3 change : 7.92 minutes

The gain in performance depends on several factors (max casts, number of loaded boxes, etc). For instance, with smaller ellipse (i.e. less wmo boxes to probe), the 2.3 gain will be much smaller. This recorded times gives a rough estimate.

Output plots were compared to ensure integrity.

[cabanesc]

• I have tested the code using wmo 6902865 (North Atlanic) until cycle 112, with the default ow_config.txt from the master branch and the latest climatology to date (CTD (2024v01) and Argo (2023v03)).
  The gain in performance is in line with the one observed for float 3902007
  initial execution : 28.17 minutes
  3.3 change : 13.12 minutes

• I have tested the integrity by comparing mapped salinities from the initial execution and the 3.3 execution. Mapped salinity should be the same as the rng() function in find_best_hist(l.49) ensures that the function ran() produces the same sequence of random numbers on each run.

However, the difference between mapped salinities is not zero on some cycles.

I have changed line 270 in update_salinity_mapping.m:

% la_bhist_Z = la_grid_Z(index);
la_bhist_Z = la_grid_Z(index_ini);

With this change, the mapped salinities are now the same for the initial and 3.3 runs

• I have then tested updating the mapping of float 6902865 up to cycle 120 (using the previous mapped files up to cycle 112). The mapped salinities are the same for the initial and 3.3 runs.

• I have also tested the code using wmo 3901901 until cycle 187, with the default ow_config.txt from the master branch and the latest climatology to date (CTD (2024v01) and Argo (2023v03)).

This float is in the South Atlantic and is crossing the 0-360 degree mark.
initial execution : 32.72 minutes
3.3 change : 14.02 minutes
Again, the mapped salinities are the same for the initial and 3.3 runs.

[delphinedobler]

Thanks Cécile, I've changed the PR15 and PR16 to include the necessary correction.

[cabanesc]

changes are included in PR #16. Closing without merge.