[MLIR][OpenMP] Independent omp.loop_nest translation

This patch refactors the process of translating loop wrappers and
`omp.loop_nest` operations so that both are independent. This makes all these
loop-related operations be handled using the same process, instead of the
previous approach where `omp.loop_nest`s were handled directly as part of their
associated loop wrappers.

In order to still have access to the collapsed loop CanonicalLoopInfo
information while translating loop wrappers, this is stored as stack frames in
the `ModuleTranslation` object initialized when translating the loop itself and
available to parent loop wrappers afterwards.

Author: skatrak

mlir/include/mlir/Target/LLVMIR/ModuleTranslation.h

@@ -290,13 +290,12 @@ class ModuleTranslation {
   /// Calls `callback` for every ModuleTranslation stack frame of type `T`
   /// starting from the top of the stack.
   template <typename T>
-  WalkResult
-  stackWalk(llvm::function_ref<WalkResult(const T &)> callback) const {
+  WalkResult stackWalk(llvm::function_ref<WalkResult(T &)> callback) {
     static_assert(std::is_base_of<StackFrame, T>::value,
                   "expected T derived from StackFrame");
     if (!callback)
       return WalkResult::skip();
-    for (const std::unique_ptr<StackFrame> &frame : llvm::reverse(stack)) {
+    for (std::unique_ptr<StackFrame> &frame : llvm::reverse(stack)) {
       if (T *ptr = dyn_cast_or_null<T>(frame.get())) {
         WalkResult result = callback(*ptr);
         if (result.wasInterrupted())