Datasets: Build upon set_ds_frequency.py to update all necessarry attributes in .zarr

[mateuszmatu]

Currently, there are three things we have to change manually after creating a .zarr file with anemoi-datasets:

• Change frequency from "1h" to 1 (already done in set_ds_frequency.py)

• There are two places where resolution is defined in .zarr. In the first location resolution is set to "o96", whereas in the second location it is set to "null". "null" has to be changed to "o96" (or whatever the resolution is set to).

• When creating a dataset with only one variable, e.g. the surface mask (at z=0), the variable name is set to "" in .zattrs of the .zarr file. This means that anemoi-training is not able to extract the variable, as it does not have a name. Add functionality to set_ds_frequency.py to handle this.
  For a variable named "land_binary_mask_0" the changes would look like this:
  "variables": [ "variables": [
  "" ------> "land_binary_mask_0"
  ] ]

And:
"variables_metadata": { "variables_metadata":{
"": { "land_binary_mask_0":{
"mars": { "mars":{
"date": "20240101", "date": "20240101"
"step": 0, "step": 0,
"time": "0000", ------> "time": "0000",
"valid_datetime": "2024-01-01T00:00:00", "valid_datetime": "2024-01-01T00:00:00",
"variable": "land_binary_mask" "variable": "land_binary_mask"
} }
} }
}, },

• Rename set_ds_frequency.py to something clever? As it will do more than just change frequency now.
• See how this issue relates to Datasets: Make one script which does ALL preprocessing up until training step #7

[mateuszmatu]

For some reason it seems that I didn't have to change variable name for "h" in "h_norkyst_2024.zarr" manually, even though there is only a single variable included here (like had to be done with the mask dataset). Strange.
A check to see if all variables are named, and then add correct names should still be added to set_ds_frequency.py.