fix(schedule): Inline non-output consumers when padding in TileWithTe…

…nsorIntrin

In the TileWithTensorIntrin function, modified the inlining behavior of consumer blocks. Now, when padding is applied, the function inlines only non-output consumer blocks. This ensures that the padding and inlining process is correctly handled for both producers and consumers.

Changes:
- Added a check to ensure only non-output consumer blocks are inlined using tir::IsOutputBlock.
- Updated the loop iterating over consumers to include the new check.

This fix addresses issues where output blocks were being inappropriately inlined, maintaining the correct block shapes and dependencies.

src/tir/schedule/transform.cc

@@ -340,7 +340,9 @@ Optional<LoopRV> TileWithTensorIntrin(const tir::Schedule& sch, const tir::Block
     }
     auto consumers = sch->GetConsumers(block_rv);
     for (const auto& consumer : consumers) {
-      sch->ComputeInline(consumer);
+      auto sref = sch->GetSRef(consumer);
+      if (!tir::IsOutputBlock(sch->state(), sref, tir::GetScopeRoot(sch->state(), sref, true)))
+        sch->ComputeInline(consumer);
     }
   }
   // Construct a mapping from tir loops back to LoopRVs