GPU-accelerated graph algorithm primitives for RingKernel.

This crate provides high-performance graph algorithms optimized for parallel execution on GPUs. It includes:

• CSR Matrix: Compressed Sparse Row format for efficient graph storage
• BFS: Parallel breadth-first search with multiple source support
• SCC: Strongly connected components via forward-backward algorithm
• Union-Find: Parallel disjoint set data structure
• SpMV: Sparse matrix-vector multiplication

Example

use ringkernel_graph::{CsrMatrix, bfs_parallel, NodeId};

// Build adjacency list: 0 -> 1 -> 2
let matrix = CsrMatrix::from_edges(3, &[(0, 1), (1, 2)]);

// Run BFS from node 0
let distances = bfs_parallel(&matrix, &[NodeId(0)]).await?;
assert_eq!(distances[0].0, 0);  // Distance to self
assert_eq!(distances[1].0, 1);  // Distance to node 1
assert_eq!(distances[2].0, 2);  // Distance to node 2