//! Sparse tensor operations: COO/CSR format conversions and sparse BLAS routines.
//!
//! # Zero-copy allocation policy
//! Functions that write every output element use `alloc_on` (uninitialized
//! allocation); functions that produce a dense output from a *sparse* source
//! (e.g. `coo_to_dense`, `csr_to_dense`) use `zeros_on` because un-referenced
//! positions must remain zero.
//!
//! | Function | Output alloc | Reason |
//! |---------------------------|--------------|--------------------------------------|
//! | `dense_to_coo` | `alloc_on` | writes every indices/values slot |
//! | `coo_to_dense` | `zeros_on` | sparse positions remain 0 |
//! | `coo_to_csr` | `alloc_on` | writes all nnz + row offsets |
//! | `csr_to_dense` | `zeros_on` | sparse positions remain 0 |
//! | `spmv` | `alloc_on` | writes every output row |
//! | `spmm` | `alloc_on` | writes every (row, col) pair |
//! | `spmm_backward_values` | `alloc_on` | writes every nnz gradient |
//! | `spmm_backward_dense` | `alloc_on` | writes every (k, n) pair |
// ── Sparse Tensor Operations Module ──
/// Sparse format conversions (COO/CSR/dense interconversion).
/// Sparse matrix operations (SpMM, SpMV, backward passes).
pub use ;
pub use ;