docs(design): SSH transport async I/O evaluation (#1593)

[oferchen]

Summary

• Expand the brief docs/design/async-runtime-ssh-eval.md into a full evaluation under docs/design/ssh-transport-async-io-eval.md for issue CI: remove unsupported windows aarch64 target #1593.
• Cover the questions prior async-SSH docs left open: tokio runtime flavour (multi-thread vs current-thread), bridge construction (tokio::process stdio vs AsyncFd over raw FD vs russh::ChannelStream), cost in RSS / scheduler footprint / blocking-vs-async learning curve, latency-sensitive read sites inside the SSH protocol (handshake greeting, multiplex headers, message dispatch), quantified workload predictions, and a five-step implementation sequencing with explicit trigger conditions for default promotion.
• Recommend a hybrid: shared rt-multi-thread runtime in the daemon (composes with rayon via spawn_blocking); per-invocation current_thread runtime in the CLI (negligible RSS overhead, sync engine untouched). Embedded path uses russh::ChannelStream; subprocess path uses tokio::process::Child stdio; AsyncFd raw-FD bridge retained as a documented escape hatch. Defer default promotion until five trigger conditions hold.

Test plan

• Docs-only change; no code paths affected.
• cargo fmt --all -- --check passes.
• Cross-references to sibling docs (async-migration-plan.md, async-ssh-transport.md, async-ssh-pipe-wrapper.md, async-ssh-evaluation.md, tokio-spawn-blocking-rayon.md) verified.
• Forward reference added to async-runtime-ssh-eval.md so the runtime-survey doc points at this evaluation.