dpnp.sum with "keepsdims" returns incorrect shape in case one or more axis is equal to 0

```python
import dpnp
import numpy

a = dpnp.ones((0, 10))
b = a.sum(keepdims=True)
print("dpnp shape", b.shape)

a_np = numpy.ones((0, 10))
b_np = a_np.sum(keepdims=True)
print("numpy shape", b_np.shape)
```

result:
```
dpnp shape ()
numpy shape (1, 1)
```