#9955 fix(tui): flush local Windows SSH replies

Summary

• flush each SSH reply chunk at the local Windows bridge boundary
• reuse the same carrier copy helper as the remote relay boundary

Root cause

The SSH carrier is long-lived. The local bridge copied ssh.exe stdout into the local socket and flushed only after EOF, so socket clients timed out before replies became visible.

Proof

• failing base behavior: https://github.com/manaflow-ai/cmux/actions/runs/31450353929/job/93653304334
• exact-head full hosted gate: pending
• speculative merge gate: pending

No local Cargo, Rust, Zig, or Xcode command was run.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

---

[!NOTE]
Low Risk
Localized Windows SSH I/O change reusing an existing flush-per-chunk helper; no auth or data-model changes.

Overview
Fixes the local Windows SSH bridge so ssh.exe stdout is forwarded to the AF_UNIX client with per-chunk flush, not only after the stream ends.

In proxy_local_connection_over_ssh, download from OpenSSH stdout now uses copy_windows_carrier_download instead of io::copy plus a single end-of-stream flush—the same path already used for the remote relay and stdio carrier boundaries.

Why: the SSH carrier stays open; buffering until EOF made local socket clients time out before replies appeared.

^{Reviewed by Cursor Bugbot for commit 60bc346904b3fe98f69cda7578519197d0876d42. Bugbot is set up for automated code reviews on this repo. Configure here.}

---

Summary by cubic

Flush SSH stdout to the local Windows socket incrementally using the shared copy_windows_carrier_download, aligning the local bridge with the remote relay behavior. Replies stream as they arrive instead of waiting for EOF, preventing client timeouts.

• Bug Fixes
  • Replaced io::copy + manual flush with copy_windows_carrier_download to write and flush each chunk.