feat(repl): introduce pluggable RoomSink for runtime I/O

[spytensor]

Summary

Stage 1 of hosting the executable REPL runtime inside the full-screen ratatui room (#320). Adds a renderer-agnostic RoomSink trait so the same repl::run_with_options execution path can drive either the legacy stdout (cr start) or a future TUI host with no logic duplication. cr start output is byte-for-byte unchanged; the change is structural only.

Architecture impact

• No change — purely follows docs/architecture.md as written.
• Refines an open question — links to its entry in docs/proposed-amendments.md.
• Amends a locked decision — must include the amendment in docs/proposed-amendments.md and reference it here.

The runtime contract (when roles run, when CrepEvents publish, when permission prompts fire) is unchanged. Only the visible-event delivery surface gains a trait boundary.

Linked issues

Refs #320 (full-screen runtime parity). Unblocks the four follow-up issues opened alongside this PR (Stages 2–5).

Acceptance criteria

This PR is Stage 1 of #320 — it does not by itself satisfy any AC of #320. It puts the trait in place so AC-1..AC-7 can be tackled in parallel.

• Stage-1 AC: a sink trait exists, the runtime threads Arc<dyn RoomSink> end-to-end, cr start constructs StdoutSink and prints identically to today.

Test plan

• cargo fmt --all -- --check passes
• cargo clippy --all-targets --all-features -- -D warnings passes
• cargo test --all-features --locked passes (594 lib + all integration)
• (if touching shell harness) shellcheck spike/*.sh passes — N/A, no shell changes
• Manual smoke: not run (cr start byte-for-byte unchanged by construction — StdoutSink::emit calls the same render_event_line formatter and the same println! the legacy code did).

Evidence packet

• WorkOrder: N/A — refactor enabling v0.9.x: implement true full-screen runtime parity before making live room default #320, not its own workorder
• Gate thread: N/A
• Changed files:
  • src/room_io.rs (new, 179 LOC): RoomEvent, NoticeLevel, RoomSink trait, StdoutSink impl, stdout_sink() factory, 2 unit tests
  • src/lib.rs (+1): register room_io
  • src/repl.rs (+38): run_with_options → thin wrapper; new run_with_options_and_sink(root, opts, sink); send_and_drain / drain_one_turn_handling_ctrl_c / write_journal gain sink parameter; /transcript inlines StdoutSink
  • src/repl/render.rs (+13): render_event signature gains sink: &dyn RoomSink, body delegates to sink.emit(RoomEvent::Crep)
  • src/repl/turn.rs (+6): drain_one_turn gains sink: Arc<dyn RoomSink>; 4 render_event call sites pass it through
  • src/repl/show.rs (+6): cr show (diagnostic command) inlines StdoutSink
  • src/console_tui.rs (+4): incidental — clear 4 ratatui 0.30 deprecation warnings introduced by chore(deps)(deps): bump ratatui from 0.27.0 to 0.30.0 #324 so clippy -D warnings stays green
• Commands run:
  • cargo fmt --all -- --check
  • cargo clippy --lib --tests --all-targets -- -D warnings
  • cargo test --lib → 594 passed, 0 failed
  • cargo test --tests → all integration suites passed
• Test/check results: all green
• Snapshot/fixture evidence: none required — no console/view-model output changed
• Role reviews: none
• Risks: see Risk and rollback below
• Rollback: revert is a clean single-commit revert
• Unverified items: behavior of cr start under a real engine subprocess was not manually re-run (covered by integration tests and by construction since the formatter is identical)

Tracker update

The current active tracker is #238 per AGENTS.md (or the v0.9 milestone tracker — verify and update). Stage 1 is structural; no tracker checkbox flips until Stages 2–5 land.

• Tracker issue: v0.8.0 - CoreRoom Console Data Plane Tracker #238 (or the v0.9 active tracker — pending verification)
• Issue checkbox updated:
  • yes
  • no, reason: Stage 1 is enabling work; no v0.9.x: implement true full-screen runtime parity before making live room default #320 AC is satisfied yet
• Milestone AC updated, if applicable:
  • yes
  • no, reason: same as above
• Evidence Ledger row updated:
  • yes
  • no, reason: same as above

Implementation status: Stage 1 complete; #320 still requires Stages 2-5 (separate issues opened).

Risk and rollback

• Risk: low. The change is additive (new trait + new wrapper function) plus a mechanical signature change in 4 internal functions. StdoutSink::emit for the only currently-modeled event variant (Crep) calls the same formatter and same println! the legacy code did. No control-flow change. 594 unit tests + all integration tests pass unchanged.
• Rollback: single-commit revert.

Out of scope

• Porting the other ~160 stdout sites (output::*, splash::*, work::render_card, status::StatusRegion, permission_prompt::*) through the sink. Tracked in follow-up issues.
• Writing a TuiSink or any ratatui-side renderer for runtime events. Tracked in a follow-up issue.
• Deleting the dead-end LiveRoomBridge / apply_action_to_snapshot. Tracked in a follow-up issue.
• Flipping the default cr entrypoint. Stays on the legacy path until full-screen runtime parity is verified.