reimplement HIP_INIT as hip_impl::hip_init(), add hip_init() to some of the inlined API (#966)

* reimplement HIP_INIT as a function, expose it as hip_impl::hip_init()
so that it could be called from hipLaunchKernelGGL and other inlined
HIP functions

* Don't call hip_init from ihipPreLaunchKernel

src/hip_clang.cpp

@@ -32,7 +32,7 @@ THE SOFTWARE.
 extern "C" std::vector<hipModule_t>*
 __hipRegisterFatBinary(const void* data)
 {
-  HIP_INIT();
+  hip_impl::hip_init();
 
   tprintf(DB_FB, "Enter __hipRegisterFatBinary(%p)\n", data);
   const __CudaFatBinaryWrapper* fbwrapper = reinterpret_cast<const __CudaFatBinaryWrapper*>(data);