NDArray.expand_dims does not correctly increase the reference count #478
Description
As a consequence, the underlying NDArray object may disappear, and the new view will not be able to access the underlying data because it has been freed already. Here it is a simple code snippet reproducing the issue:
In [1]: import blosc2
In [2]: a = blosc2.arange(10)
In [3]: a = blosc2.expand_dims(a)
In [4]: a[0]
---------------------------------------------------------------------------
RuntimeError Traceback (most recent call last)
Cell In[4], line 1
----> 1 a[0]
File ~/blosc/python-blosc2/src/blosc2/ndarray.py:1782, in NDArray.__getitem__(self, key)
1780 # Create the array to store the result
1781 arr = np.empty(shape, dtype=self.dtype)
-> 1782 nparr = super().get_slice_numpy(arr, (start, stop))
1783 if step != (1,) * self.ndim: # TODO: optimise to work like __setitem__ for non-unit steps
1784 # have to make step refer to sliced dims (which will be less if ints present)
1785 slice_ = tuple(slice(None, None, st) for st, m in zip(step, nm_, strict=True) if m)
File ~/blosc/python-blosc2/src/blosc2/blosc2_ext.pyx:2426, in blosc2.blosc2_ext.NDArray.get_slice_numpy()
File ~/blosc/python-blosc2/src/blosc2/blosc2_ext.pyx:2248, in blosc2.blosc2_ext._check_rc()
RuntimeError: Error while getting the buffer
The solution imply re-thinking views, and how to use them from C/Python.