Step away from "type specifiers" introduced in xcube 0.6

[forman]

Is your feature request related to a problem? Please describe.

The "type specifiers" (class xcube.core.store.TypeSpecifier) concept introduced in xcube 0.6 is no longer required in the light of the data cube convention, see #488. It complicates the data store API, where simple data type names are sufficient and less confusing.

Describe the solution you'd like

We can now simply use a parameter data_type: DataTypeLike where type_specifier: str was expected, wIth

• DataType = Union[xr.Dataset, xcube.core.multilevel.MultiLevelDataset, gpd.GeoDataFrame, Any] and
• DataTypeLike = Union[str, DataType].

Function implementation can use a new utility function normalize_data_type(data_type: DataTypeLike) -> DataType to get the actual type.

When users really need a cube in terms of (time, ..., y, x) dimensions, they use the new dataset’s cube subset function xcube.core.normalize.get_dataset_cube_subset() -> Tuple[xr.Dataset, GridMapping] or simply xr.Dataset.xcube.xcube.to_cube() (the xcube property is an xarray dataset extension), therefore there is no more need to use type specifiers with flags, e.g. "dataset[cube]":

dataset = store.open_data(…, data_type=xr.Dataset)
cube, grid_mapping = dataset.xcube.to_cube()

Same for multi-level datasets:

ml_dataset, grid_mapping = dataset.xcube.to_ml_dataset(num_levels=5, tile_size=512)

Original type specifier names with flags now become simple data type names:

• From "dataset[cube]" to "dataset" or "xarray.Dataset"
• From "dataset[multilevel]" and "dataset[multilevel, cube]" to "ml_dataset" or "xcube.core.multilevel.MultiLevelDataset"
• Allow "geodataframe" and also "geopandas.GeoDataFrame"