[llvm] [mlir] [mlir][gpu] Change GPU modules to globals (PR #135478)

[Mehdi Amini via llvm-commits]

joker-eph wrote:

I think we should move it to `createKernelLaunch` because this is a layering problem: the choice of the runtime should be part of the lowering of the GPU dialect, it is meant to be customizable for other runtime.
The move to the translation layer makes it non longer a customizable lowering thing, but something that requires non-trivial workarounds now (basically you need to inject an empty `mgpuModuleLoad` in the LLVM module to "cancel" this added code).

https://github.com/llvm/llvm-project/pull/135478