Support nested dicts in DataTree.from_dict() and to_dict()

[shoyer]

What is your issue?

It would be nice to natively support nested data in from_dict() and to_dict(), rather only than the current "flat" format, e.g., DataTree.from_dict({'a': {'b': {'c': None}}}) instead of DataTree.from_dict({'/a/b/c': None}).

For from_dict(), we could support this by automatically expanding dictionary arguments.

For to_dict(), we could add a new argument nested=True for indicating that a nested dict of names should be returned instead of the current flat structure.

[shoyer]

Supporting nested dicts in from_dict would be particularly nice paired with the support for DataArray convertible arguments from #10658.

Instead of writing something like:

        >>> dt = DataTree.from_dict(
        ...     {
        ...         "/": Dataset(coords={"time": [1, 2, 3]}),
        ...         "/ocean": Dataset(
        ...             {
        ...                 "temperature": ("time", [4, 5, 6]),
        ...                 "salinity": ("time", [7, 8, 9]),
        ...             }
        ...         ),
        ...         "/atmosphere": Dataset(
        ...             {
        ...                 "temperature": ("time", [2, 3, 4]),
        ...                 "humidity": ("time", [4, 5, 6]),
        ...             }
        ...         ),
        ...     }
        ... )

You could simply drop the Dataset() constructors from most nodes, e.g.,

        >>> dt = DataTree.from_dict(
        ...     {
        ...         "/": Dataset(coords={"time": [1, 2, 3]}),
        ...         "/ocean": {
        ...             "temperature": ("time", [4, 5, 6]),
        ...             "salinity": ("time", [7, 8, 9]),
        ...         },
        ...         "/atmosphere": {
        ...             "temperature": ("time", [2, 3, 4]),
        ...             "humidity": ("time", [4, 5, 6]),
        ...         },
        ...     }
        ... )