test(apprt-gtk): replay mouse input and verify real-world TUIs

The GTK mouse handlers now delegate to *_impl methods taking button and
modifier state, so the real controllers and the debug input replay share
one code path. The replay hook gained mouse-move/press/release, click,
drag, scroll, resize, clipboard-set/dump, windows-dump and
window-screenshot steps (replayed presses bypass the focus-transfer
first-click rule since the replay window isn't compositor-focused).

Verified in the real app on Wayland (docs/rewrite/notes/verification-
V3.md): nvim (colors, status line, cursor shapes, alt-screen restore),
htop (rendering, mouse-click column sorting), less (alternate scroll),
tmux (splits, status bar), drag/double/triple-click selection into
PRIMARY, clipboard copy, context menu, scrollback and reflow on resize,
CJK/emoji/ZWJ/nerd-font widths, kitty graphics images and the unsafe
paste confirmation dialog. No product bugs were found.

Test Plan:
- cargo test -p ghostty-apprt-gtk (incl. parse_mouse_steps)

docs/rewrite/notes/verification-V3.md

@@ -0,0 +1,43 @@
+# Verification V3 (real GTK app, real programs, mouse input)
+
+## Replay hook extensions (`crates/ghostty-apprt-gtk/src/debug_input.rs`, `class/surface.rs`)
+New steps: `mouse-move x y [mods]`, `mouse-press|mouse-release|click <btn> [mods]`, `drag x1 y1 x2 y2 [mods]`,
+`scroll dx dy`, `resize w h`, `clipboard-set`, `clipboard-dump` (also logs context-menu visibility),
+`windows-dump` (toplevels + widget tree, shows AdwDialogs), `window-screenshot` (GTK render of the whole
+window incl. headerbar/dialogs; the GL area content may differ from `screenshot`).
+The GTK handlers `mouse_down/mouse_up/mouse_motion/scroll_vertical` were split into `*_impl` methods
+(taking button/modifier state instead of a `GdkEvent`) so the replay runs the exact same code as the
+gesture controllers. The replay passes `assume_focus=true` for presses: the replayed window is not
+compositor-focused, and the "first click only transfers focus" suppression would otherwise eat every click.
+Parser unit test `parse_mouse_steps` added.
+
+## Scenarios (GTK, Wayland, release profile)
+| Scenario | Result |
+|---|---|
+| nvim (-e nvim file.rs): syntax colors, ligatures, status line, `:q` | PASS. Insert mode bar cursor OK. |
+| nvim from shell, `:q`: alternate screen restore | PASS (prompt + prior output intact). |
+| htop: meters, colors, box/bars, F10 | PASS. Mouse click on MEM% header sorts (mouse reporting; clicking twice toggles asc/desc). F10 quits. |
+| less on 300 lines: wheel in alt screen, `q` | PASS. 5 wheel ticks -> 15 lines (3 arrow keys per tick, DECSET 1007), `q` restores. |
+| tmux: start, ctrl+b %, status bar, exit | PASS. |
+| shell drag selection | PASS (highlight; text lands in PRIMARY via copy-on-select). |
+| double-click word / triple-click line | PASS ("beta"; whole line). |
+| ctrl+shift+c | PASS (CLIPBOARD gets the selection). |
+| right-click | PASS: context popover becomes visible (`context menu visible = true`). |
+| wheel scrollback (10 ticks = 30 lines) | PASS. |
+| long wrapped line + window resize (shrink, grow) | PASS: reflow correct both ways, viewport preserved. |
+| unicode: CJK, emoji, ZWJ family, combining, fullwidth, powerline/nerd glyphs, box drawing | PASS: widths/alignment correct. |
+| kitty graphics (PNG, a=T, c=8,r=4) | PASS: image rendered at the right cell size, text flows after it. |
+| multi-line paste into `cat` (no bracketed paste) | PASS: core rejects as unsafe, `GhosttyClipboardConfirmationDialog` is created and mapped (widget tree dump). The dialog isn't visible in the render-texture screenshot, only verified structurally. |
+| Raw mouse bytes (SGR 1006, button 1/3, wheel) | PASS: `ESC[<0;10;5M/m`, `ESC[<2;..`, `ESC[<64;..` x3 per tick. |
+| Kitty keyboard protocol (flags 1/31) from `key` steps | PASS (CSI u with text/release events). |
+
+## Bugs
+No product bugs found in this pass. Observations:
+- Release profile keeps `debug-assertions = true` (workspace Cargo.toml), so the "debug build" banner shows; use `--profile production` to get the real release look.
+- `screenshot` returns the renderer's last presented frame; right after input it can be one frame stale. Use >= 1 s waits.
+- Replay limits: `key <printable>` bypasses the IM; `text` commits bypass key events (plain bytes even under kitty flags "report all keys"); window is not compositor-focused so cursor is hollow and clicks need `assume_focus`.
+- nvim leaves swap files when the replay is `quit` mid-edit; a later nvim of the same file prompts (E325) and swallows keystrokes. This looked like a bug at first but is nvim behaviour (use `nvim -n`).
+
+## Remaining
+- Confirm/deny of the paste dialog by interacting with it not exercised; IME via real input method not exercised.
+- Horizontal scroll (tab switching) not replayed.