This change implements memory-mapped views for index readers and sparse matrices, enabling shared page-cache access. It also includes performance analysis confirming that the default chunk size cap negatively impacts parallelism for small inputs.