Use `xr.Dataset` in krill data ingestion

[leewujung]

This is similar to #369 but more specific to functions in the krill data ingestion steps.

This applies to outputs of aggregate_cell.

Made this a separate issue because #369 is for functions in the hake bio data processing pipeline.

[leewujung]

Just adding a clarification note from our last meeting: the upgrade to using xr.Dataset for storing data is only for consolidating the inputs (before going into inversion) into a 3D Sv dataarray (dimensions: depth x interval x frequency).

For the inversion step, the ops would be to slice out a particular depth layer or to sum over all depth layers from the 3D Sv dataarray (depth x interval x frequency) to obtain a 2D dataframe containing Sv for each interval (row: interval, column: frequency). The inversion will still be done on the dataframe, so no change needed there.