feat(tui): trial an anchored composer in fullscreen mode on nightly

[me2seeks]

Summary

Trial the existing fullscreen TUI mode on the nightly channel so the composer can remain anchored while users read earlier transcript content.

This issue follows Discussion #3879. The pinned @earendil-works/pi-tui@0.84.2 dependency already exposes a fullscreen/alternate-screen path, so this should be evaluated as a scoped product experiment rather than a renderer rewrite.

Experiment scope

• Run the fullscreen TUI path in the nightly channel.
• Keep the composer available while the transcript viewport scrolls independently.
• Preserve the user's reading position when new model or tool output arrives.
• Show a clear unread/new-output indication while the user is away from the bottom.
• Exercise application-owned mouse scrolling, text selection, copying, transcript search, and links.

Evaluation questions

• Is returning to the bottom before typing a meaningful problem in real use?
• Does the anchored composer materially improve reading-while-working?
• Is application-owned selection and copying acceptable compared with native terminal behavior?
• Is losing terminal-native scrollback during and after the session acceptable?
• Should the status line remain anchored with the composer, collapse, or scroll with the transcript?
• What should happen when a user starts typing while viewing older content?

Non-goals

• Do not make fullscreen the stable default as part of this issue.
• Do not add a permanent user-facing mode toggle before the nightly experiment produces evidence.
• Do not fork or rewrite the upstream renderer unless the existing fullscreen path proves insufficient.

Exit criteria

Document nightly feedback and observed behavior for reading position, unread indication, selection/copy, terminal history, resizing, and small terminals. Use that evidence to decide whether to stabilize, revise, or drop the mode.

AI assistance disclosure: Drafted and published with Codex at the author's direction.

[abhinav-phi]

I’d like to take this up. I can run and evaluate the fullscreen TUI experiment, focusing on the anchored composer, independent transcript scrolling, unread/new-output indication, selection/copy, resizing, and small-terminal behavior, and document the findings against the exit criteria.

Could you please assign this to me?

/assign