ENH: xarray-like .loc / .sel for named model vars

[gt-lc-0000]

Context

The idea for this new feature is to make it easier to slice named model variables.
A model variable is a symbolic TensorVariable, which makes it hard to interact with.
However, with .coords and .RV_dims, we have the necessary information to perform labeled slicing.

Hopefully the use case isn't too nice. For my case, it's useful for nested hierarchical models.
Models with multi-dimensional variables would also be easier to work with, if slicing / re-indexing is needed.

Before

def sel(
    model: "pm.Model", varname: str, indexers: dict[str, typing.Union[str, list[str]]]
) -> "pytensor.tensor.var.TensorVariable":
    """Like xarray.loc / .sel for named variables in a ``pymc.Model`` instance

    Parameters
    ----------
    model : pm.Model
    varname : str
    indexers : dict[str, str or list[str]]

    Returns
    -------
    pytensor.tensor.var.TensorVariable
        sliced model variable
    """
    try:
        dims = model.RV_dims[varname]
    except KeyError as err:
        raise ValueError(f"'{varname}' has no dimensions") from err

    def process_dim(dim):
        try:
            vals = indexers[dim]
        except KeyError:
            return slice(None)
        else:
            return (
                model.coords[dim].index(vals)
                if isinstance(vals, str)
                else [model.coords[dim].index(val) for val in vals]
            )

    return model[varname][tuple(process_dim(dim) for dim in dims)]

After

# expected to only work for multi-dim vars, one variable at a time
model.loc[{dim1: [val1], dim2: val2}, varname]