[clang] [CIR] Implement FlattenCFG for coroutine AwaitOp, CoroBodyOp, and CoReturnOp (PR #203802)

[Bruno Cardoso Lopes via cfe-commits]

================
@@ -1805,11 +1813,253 @@ class CIRTryOpFlattening : public mlir::OpRewritePattern<cir::TryOp> {
   }
 };
 
+static mlir::Block *getOrCreateBlockForSuspendPoint(
+    cir::FuncOp funcOp, mlir::PatternRewriter &rewriter, mlir::Location loc) {
+  mlir::Block &entryBlock = funcOp.getBody().front();
+
+  auto it = llvm::find_if(entryBlock, [](auto &op) {
+    return mlir::isa<AllocaOp>(&op) &&
+           mlir::cast<AllocaOp>(&op).getCoroutineSuspendPoint();
+  });
+
+  assert(it->hasOneUse() &&
+         "coroutine suspend point alloca must have exactly one use");
+  auto storeOp = cast<cir::StoreOp>(*it->getUses().begin()->getOwner());
+  auto suspendPoint = cast<cir::ConstantOp>(storeOp.getValue().getDefiningOp());
+  mlir::Block *suspendBlock = suspendPoint->getBlock();
----------------
bcardosolopes wrote:

Even setting aside the agreed redesign, note this derives the destination from the **constant's** block (`suspendPoint->getBlock()`), not the store's. Constants are the most mobile thing in the IR (hoisted to entry, CSE-merged across the whole function), so any other `cir.const 1` can relocate or merge this one and silently repoint the suspend destination, which is exactly the fragility that motivates dropping the positional beacon. If any interim version survives, `storeOp->getBlock()` would at least be less mobile. (Also: the stored value `1` is currently vestigial, nothing reads it, only the position matters — so today the mechanism only uses the half that canonicalization breaks.)

https://github.com/llvm/llvm-project/pull/203802