Allow geom_bar() as a marginal layer

Example:

```python
data = {
'x': ["a", "a", "b", "b"],
'y': ["f", "g", "g", "g"],
}
ggplot(data, aes('x', 'y')) + \
geom_count() + \
ggmarginal('r', layer=geom_bar())
```

Output:

> Internal error: IllegalStateException : Not a collections of Double(s)

Can be fixed by adding `orientation='y'` into the `geom_bar()`.

It seems that the ability to add discrete geoms as marginal layers was overlooked, but with the introduction of `geom_count()` it makes sense.

Start by running the provided Python example with geom_count() and ggmarginal('r', layer=geom_bar()). Trace the marginal-layer handling for geom_bar and its orientation inference. Done means the example works without explicitly setting orientation='y' and still produces the expected marginal bar layer.

Written by the indexing model from the issue text.