Update MEPS npy example

[leifdenby]

I am just writing some notes here with suggestions for what could be done to update the MEPS Npy example:

• Add README based on "Format of data directory" from v0.1.0 of neural-lam README https://github.com/mllam/neural-lam/tree/v0.1.0?tab=readme-ov-file#format-of-data-directory
• Move information about step_length (in number of hours), number of timesteps in forecast and number of ensemble members into config (https://github.com/mllam/neural-lam/blob/main/neural_lam/weather_dataset.py#L58, https://github.com/leifdenby/neural-lam/blob/feat/datastores/neural_lam/datastore/npyfiles/store.py#L160)

[sadamov]

Since we already have this issue here, I am adding the point that we wanted to make sure that the timesteps covered by this test-datastore match the danra test-datastore.
currently the time-coverage is:

• danra: 1990-09-03T00:00 - 1990-09-09T00:00
• meps: 2022-04-01 - 2022-09-15 (with gaps)

It would be fantastic to have both datasets cover roughly the same time-period (can be identical for train/val/test). For danra you could keep the same time-steps.
For meps very few analysis-time steps are required, because the slicing happens along elapsed_forecast_period . So I would say 10 analysis_times, maybe.