block statements are now directly following the Block AST node rather
than a singly linked list. This had negligible impact on performance:

throughput: 72.3 MiB/s => 72.7 MiB/s

however it greatly improves the API since the statements are laid out in
a flat array in memory.