improve cuda gufunc: multiple output arguments

failing example

``` python
import numpy as np
import numba as nb

@nb.guvectorize('f8[:], f8[:], f8[:]', '()->(),()', nopython=True, target='cuda')
def foo(inp, out1, out2):
out1[0] = inp[0] * 2.3
out2[0] = inp[0] + 1

inp = np.arange(10, dtype=np.float64)
out1 = np.empty_like(inp)
out2 = inp
for _ in range(3):
foo(inp, out=(out1, out2))
print('out1', out1)
print('out2', out2)
```

Run the failing Python example with the CUDA target and inspect the gufunc handling for multiple output arguments. Confirm that repeated calls with out=(out1, out2) produce the expected separate outputs; the issue does not name a source file or test to begin with.

Written by the indexing model from the issue text.