Optional bounded Domainslib execution for ModelKit.
Requested domains include the calling domain. A value of one delegates to ModelKit's portable sequential backend without creating a Domainslib pool. Parallel execution uses O(tasks) atomic result slots, runs at most domains task bodies concurrently, preserves logical result order, and returns the error from the lowest failing task index.
inner_threads describes the maximum threads used inside one task. When it is omitted, common BLAS/OpenMP environment limits are inspected and the largest valid value is used; otherwise one is assumed. Diagnostics never mutate process environment or numerical-library settings.
create ~domains () validates a positive total domain count. Supplying inner_threads overrides environment detection for diagnostics only; the backend does not configure numerical libraries.