Certain applications (pytorch) decide that it's a great idea to distribute whole CUDA driver and link to it with DT_RPATH. This igores LD_LIBRARY_PATH.
This code defeats that evil mechanism through any means necessary
On my machine ZLUDA seems to segfault when initializing LLVM's C++ statics in Blender. Blender ships with C++ runtime. It seems that compiling C++ runtime statically fixes the issue. Might be actually unrelated.
Additionally, dtor crate on Windows seem to use a slightly dodgy method, so replace it with something more straightforward
Fixes issues pointed out in #27:
* spirv_tools-sys was build in non-test profiles
* By default ZLUDA dll has a wrong name
* We relied on third-party OpenCL installation on Windows
* We encouraged building debug configuration
* We didn't provide build information for developers (cmake, python, submodules)