JIT issues when using jupyter notebook

Hi, after updating to the latest arrayfire jupyter notebook seems to have a problem with `af.gaussian_kernel()` function. Standard `ipython` works without a problem but in notebook the call ends up with:

```
RuntimeError Traceback (most recent call last)
in ()
----> 1 lala = af.gaussian_kernel(128,128,5,5)

~/anaconda3/lib/python3.6/site-packages/arrayfire/image.py in gaussian_kernel(rows, cols, sigma_r, sigma_c)
778 safe_call(backend.get().af_gaussian_kernel(c_pointer(out.arr),
779 c_int_t(rows), c_int_t(cols),
--> 780 c_double_t(sigma_r), c_double_t(sigma_c)))
781 return out
782

~/anaconda3/lib/python3.6/site-packages/arrayfire/util.py in safe_call(af_error)
77 err_len = c_dim_t(0)
78 backend.get().af_get_last_error(c_pointer(err_str), c_pointer(err_len))
---> 79 raise RuntimeError(to_str(err_str))
80
81 def get_version():

RuntimeError: In function std::vector cuda::compileToPTX(const char*, std::string)
In file src/backend/cuda/jit.cpp:
```
I tried complete reinstall of python, cleaned any personal settings in .local .bashrc .config and installed completely new anaconda. Still whatever I do I cant get it to run

NOTE: `af.constant()` function works without problem
SYSTEM: ubuntu 17.10; CUDA 9-2; arrayfire v3.6 binary; arrayfire-python git master

Start by reproducing af.gaussian_kernel(128,128,5,5) in the reported Jupyter and IPython environments, then trace the failure through arrayfire/image.py and arrayfire/util.py into src/backend/cuda/jit.cpp. Done means the call works in Jupyter under the reported setup, with the CUDA JIT error understood or prevented.

由索引模型根据 Issue 内容生成。