Fix two test failures with CUDA 9.2 on Windows

fix docs typo

fix to make device_allocation.as_buffer() result writable

Previously the same fix has been applied to pointer_holder_base.as_buffer but not for device_allocation class

Fix documentation related to stop_profiler method.

Add get_texref() to ElementwiseKernel

See merge request !10

_new_like_me() doesn't copy strides/order

Equivalent to:
https://github.com/inducer/pyopencl/issues/58

Fix slicing with negative stride.

fix: update signature of gpuarray.reshape to match the GPUArray method

See merge request !8

fix for CUDA 40

while compiling on a machine with CUDA 40, CU_LIMIT_DEV_RUNTIME_SYNC_DEPTH triggered a non-declared error.
it seems that the version  requirement is higher than 4000

Ipow

See merge request !6