docs(design): io_uring submission from rayon worker threads (#1284)

[oferchen]

Summary

• Documents current io_uring submission topology: disk-commit thread is the sole submitter for receiver writes; per-file rings are built on the calling thread for reads and non-batched writes. Zero submissions originate from rayon workers today.
• Enumerates per-rayon-worker hazards: worker-vs-ring lifetime mismatch, fixed-file registration churn, registered-buffer pool multiplied per ring, SQPOLL kthread proliferation, work-stealing breaking FD-to-ring affinity.
• Lays out the single-submitter MPSC alternative and the bench evidence still needed (bench(fast_io): io_uring per-file vs shared ring on 100K small files (#1410) #4197, bench(fast_io): io_uring SQPOLL vs regular submission vs std::fs (#1626) #4201, bench(engine): drain_parallel alternatives at 10K/100K items (#1682) #4214) before a decision can land.

Recommendation

Defer until #2243 (per-thread rings) and #4197 (per-file vs shared ring bench) converge. Keep the disk-commit thread as the sole submitter for writes; keep per-file rings for the generator read side.

Test plan

• Pure docs change; no code touched
• cargo fmt --all clean
• File path: docs/design/iouring-rayon-submission.md