Axis.startswith,endswith,matches should return an IGroup instead of an LGroup

Because they have a better behaviour in the presence of duplicate labels

```python
>>> a = Axis('a=a0,a0,a1')
>>> arr = ndrange(a)
>>> arr
a  a0  a0  a1
    0   1   2
>>> arr[a.startswith('a')]
a  a0  a0  a1
    1   1   2
```