clspv should use LLVM atomic instructions

clspv currently replaces OpenCL atomic builtins with SPIR-V builtins. Instead we should use llvm atomic instructions throughout the flow and translate them in SPIRVProducerPass. Currently some of the builtins are handled this way. Increment, decrement and compare exchange need updated.

Other issues with atomics in clspv: 1.2 opencl atomics should be relaxed and map to Monotonic ordering in LLVM.

All OpenCL atomic builtins claim the pointer is volatile. So at least through the clspv flow atomic instructions should get marked as volatile.