Only the analysis part of the interface is implemented. The bufferization itself is performed by the SparseTensorConversion pass. Differential Revision: https://reviews.llvm.org/D128138