[PIPELINING] Fix stage for the local_load in the TMA pipelining (#5365)

`local_load` should be in the same stage that the `subview` that it is
using.

Author: pawelszczerbuk

lib/Dialect/TritonGPU/Transforms/Pipeliner/MatmulLoopPipeline.cpp

@@ -272,7 +272,7 @@ createTMAAsyncCopy(scf::ForOp &forOp, tt::ExperimentalDescriptorLoadOp loadOp,
 
     builder.setInsertionPointAfter(viewLoad);
     auto sharedLoad = builder.createWithStage<ttg::LocalLoadOp>(
-        loc, stage, clusterId, loadOp.getType(),
+        loc, stageForFirstUse, clusterForFirstUse, loadOp.getType(),
         viewLoad /*,wait->getResult(0)*/);
     auto result = sharedLoad->getResults();
     loadOp->replaceAllUsesWith(result);