[mlir][sparse] fix memory leak sparse2sparse reshape

mlir/lib/Dialect/SparseTensor/Transforms/SparseTensorRewriting.cpp

ForeachOp foreachOp = rewriter.create<ForeachOp>(
             srcIndices.push_back(args[dim]);
           }
           translateIndicesArray(builder, loc, op.getReassociationIndices(),
                                 srcIndices, srcSizes, dstSizes, dstIndices);
           auto t = builder.create<InsertOp>(loc, v, reduc.front(), dstIndices);
           builder.create<sparse_tensor::YieldOp>(loc, t);
         });
     auto t = rewriter.create<LoadOp>(loc, foreachOp.getResult(0), true);
-    rewriter.replaceOpWithNewOp<ConvertOp>(op, dstTp, t);
+    auto converted = rewriter.create<ConvertOp>(loc, dstTp, t).getResult();
+    rewriter.create<DeallocTensorOp>(loc, t);
+    rewriter.replaceOp(op, converted);
     return success();
   }
 };
 
 /// Sparse rewriting rule for sparse-to-dense and dense-to-sparse reshape
 /// operator.
 template <typename ReshapeOp>
 struct ReshapeRewriter : public OpRewritePattern<ReshapeOp> {