Multiprocessing doesn't preserve data sharing of storage slices

``` python
x = torch.Storage(10)
y = x[1:-1]

#1
with open('file.t7', 'w+b') as f:
    torch.save([x, y], f)
    f.seek(0)
    a, b = torch.load(f)
# a and b no longer share the same storage

#2
q = multiprocessing.Queue()
q.put([x, y])
a, b = q.get()
# a and b no longer share the same data
```
