Introduce support for handling multiple datasets with varying spatial grids and time resolutions across the Anemoi framework. This milestone will allow users to independently control which data sources are used as model inputs and outputs and define each source's time step individually.

This is a key intermediate step toward enabling a broader range of use cases within Anemoi, such as:

• Forecasting
• Spatial downscaling
• Time interpolation
• Coupled training
• Complex multi-source learning setups (nowcasting, LAM, observations, ...)

By standardizing how heterogeneous data is managed, this milestone enhances Anemoi’s modularity and prepares the system for future support to this use cases.