Allow alternative formats for input and snapshots variables

[benbovy]

Currently the only accepted format for input and snapshots variables is nested dictionaries of processes and variables, e.g.,

in_ds = xsimlab.create_setup(
    ...
    input_vars={
        'grid': {
            'x_size': 201,
            'y_size': 101,
            'x_length': 1e5,
            'y_length': 1e5
        },
        'topography': {
            'elevation': (('y', 'x'), np.random.rand(101, 101))
        }
    },
    snapshot_vars={
        'out': {
            'topography': 'elevation'
        },
        None: {
            'grid': ('x', 'y')
        }
    }
)

Although a bit more verbose, flat dictionaries may be better in some cases (It is also consistent with the flat mapping of data variables in new returned xarray.Dataset objects):

in_ds = xsimlab.create_setup(
    ...
    input_vars={
        'grid__x_size': 201,
        'grid__y_size': 101,
        'grid__x_length': 2e5,
        'grid__y_length': 1e5,
        'topography__elevation': (('y', 'x'), np.random.rand(101, 201))
    },
    snapshot_vars={
        'out': ('topography__elevation'),
        None: ('grid__x', 'grid__y')
    }
)

We might also accept tuples, which are easier to use programmatically (this may be useful if someday "virtual groups" get implemented in xarray, see this issue comment):

in_ds = xsimlab.create_setup(
    ...
    input_vars={
        ('grid', 'x_size'): 201,
        ('grid', 'y_size'): 101,
        ('grid', 'x_length'): 2e5,
        ('grid', 'y_length'): 1e5,
        ('topography', 'elevation'):  (('y', 'x'), np.random.rand(101, 201))
    },
    snapshot_vars={
        'out': (('topography', 'elevation'),),
        None: (('grid', 'x'), ('grid', 'y'))
    }
)

If we support the three different formats, we also need to decide if we allow mixing these formats. It should be easy to implement (i.e., test if key is tuple, or <process> or <process__variable>).

[benbovy]

Closed in #33