#592-D: register GTK scroll surfaces + hit-test caches on the live path, then delete src/gtk/draw.rs

[JDonaghy]

Sub-task 4/4 of #592 — the closer. Depends on #592-A, #592-B and #592-C: src/gtk/draw.rs is the reference implementation those three port from, so it cannot be deleted until they land.

What

Two halves, in order:

1. Re-home the state the dead path was the sole writer of. Per #592, GTK's scroll and
click dispatch currently run against structures nothing live ever fills:

• engine.scroll_surfaces is written only from dead code (draw.rs:85, :644, :709).
  GTK's dispatch_scroll (mod.rs:1838) and dispatch_click (mod.rs:3577) therefore
  hit-test against a permanently empty list. TUI registers its surfaces every frame.
• Several hit-test caches are read by live GTK event handlers but written only by the dead
  path — enumerate them from GTK live render path silently drops 13 populated ScreenLayout fields (rest of #587's blast radius) #592's table and from a compile check once draw.rs is gone.

2. Delete src/gtk/draw.rs.

Why this is the payoff

src/gtk/draw.rs is 3,733 lines under a file-level #![allow(dead_code)]. Of its 29
exported pub(super) fns, zero have a live caller (verified on origin/develop
2026-08-26; the one apparent hit on draw_editor is a comment). That file-level allow is
what muted the warning that would have caught #587 and #592 years earlier — deleting the
file removes both ~3.7k lines of dead code and the mechanism that hid the bug.

Design

• Do the re-homing before the delete, and land it as its own commit, so a regression is
  bisectable to one half.
• Register scroll surfaces from the live render_content path at paint time, the way TUI
  does — one shared registration path, not a GTK-specific one.
• Remove the file-level #![allow(dead_code)] from any file that still carries it and let
  the compiler name whatever is left. Anything it flags is either a real orphan (delete) or a
  missed port (a bug in A/B/C — go back and fix it there rather than reviving draw.rs).
• If deleting draw.rs reveals a surface no one ported, stop and say so rather than
  copying the function back into mod.rs.

Acceptance criteria

• src/gtk/draw.rs no longer exists.
• No file under src/gtk/ carries a file-level #![allow(dead_code)].
• GTK scroll-wheel dispatch and click dispatch hit-test against a surface list that is
  actually populated — proven by a GtkDriver test that scrolls a scrollable surface and
  asserts the content moved, which cannot pass against an empty list.
• All thirteen GTK live render path silently drops 13 populated ScreenLayout fields (rest of #587's blast radius) #592 surfaces still paint (the A/B/C tests stay green).
• TUI behaviour unchanged.
• cargo build && cargo test EXIT=0.

Files

• src/gtk/draw.rs
• src/gtk/mod.rs
• src/gtk/click.rs
• src/gtk/testing.rs
• src/render.rs

Out of scope

• vimcode's private enum Msg bus (mod.rs:990-1319) — related debt from the same #448-C: Flip GTK main loop from Relm4 to ShellApp runner; delete Relm4 scaffolding (sub-task 3/3 of #448) #540
  migration, but a separate decision; file it if this issue makes the case clearer.
• Ctrl+V paste silently does nothing on GTK — quadraui swallows the key, vimcode has no ClipboardPaste handler #593 (Ctrl+V / UiEvent::ClipboardPaste), same root cause, different fix.

[JDonaghy]

Coordinator: Assignment Complete

Machine: dellserver
Status: done
Exit code: 0
Duration: 33m 32s
Log: /home/john/.coord/logs/c93b31700fab.log

[JDonaghy]

🧊 Pipeline row stalled

Machine: dellserver
Assignment: c93b31700fab
Issue: #672
Reason: Work done, no review ever dispatched

Work is done with test_state='passed' but no review assignment was ever dispatched for it.

Detection + surfacing only — nothing was dispatched automatically. This row's auto-loop transition already fired once and won't fire again on its own; a human should look at coord status and decide whether to dispatch a fix/review manually or drop it with coord backlog.