Module parallel 

Native parallel module for data-parallel operations.

Exports: parallel.map, parallel.filter, parallel.for_each, parallel.chunks, parallel.reduce, parallel.num_threads

Uses Rayon for thread-pool based data parallelism. The key constraint is that Shape closures (ValueWord) are not Send, so we use invoke_callable on the main thread but process pure-data operations (chunking, collecting) in parallel where possible.

For parallel.map and parallel.filter, the callback is invoked via invoke_callable on the calling thread, but the data is partitioned and reassembled using Rayon when no callback is involved.

Functions

create_parallel_module

Create the parallel module.