misc-static-initialization-cycle is very slow

After upgrading from LLVM19 to 23, we did some performance tests on a selection of large files in our codebase.
The check `misc-static-initialization-cycle` seems to take a significant part of the time.
````
===-------------------------------------------------------------------------===
clang-tidy checks profiling
===-------------------------------------------------------------------------===
Total Execution Time: 29.0469 seconds (29.0015 wall clock)

---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Name ---
2.8125 ( 10.5%) 0.0938 ( 4.3%) 2.9062 ( 10.0%) 2.8914 ( 10.0%) misc-static-initialization-cycle
````

A first check of this code points out the following code:
````
bool VisitDeclRefExpr(DeclRefExpr *DRE) override {
if (const auto *VarD = dyn_cast(DRE->getDecl())) {
if (!shouldIgnoreRef(DRE, Node->getDecl()) &&
(VarD->hasGlobalStorage() || VarD->isStaticLocal()))
Node->Uses.emplace_back(DRE, G.addNode(VarD->getCanonicalDecl()));
}
return true;
}
````
The `shouldIgnoreRef` method is very expensive, as it calls `ParentMapContext::getParents(...)` in all cases followed by some recursive looping towards the grandparents.

If I understand the code correctly, the method `shouldIgnoreRef` has no side effects. So by swapping the arguments of `&&`, several calls into `shouldIgnoreRef` could be prevented as a quick win.

Start at misc-static-initialization-cycle's VisitDeclRefExpr and inspect shouldIgnoreRef, especially its ParentMapContext::getParents calls and recursive parent traversal. Check whether evaluating the global-storage or static-local condition first preserves behavior while avoiding unnecessary calls, then compare clang-tidy profiling results and ensure the check still reports the same cycles.

Written by the indexing model from the issue text.