Improvements and bug-fixes:

- Moved all constant NVRTC compiler configuration from DynamicSourceModule.add_source_nvrtc() to constructor of NvrtcCompiler
- Added support for NVRTC in-memory header files
- Added support to let the PyCUDA user create and preconfigure a NvrtcCompiler instance and pass it to add_source_nvrtc()
- Added argument "exports" to constructor of DynamicSourceModule, triggers NVRTC compilation instead of NVCC
- Fixed C++ buffer handling