fix: respect xkb keysym remaps for editing keys

[raviboth]

Why I'm filing this

I use the Colemak keyboard layout and have Caps Lock remapped to Backspace at the xkb level — this is a very common ergonomic setup (Caps Lock is unused real estate, Backspace gets used constantly, the remap puts it under a home-row finger). Every other app on my system honours the remap. In Seance the remapped key does nothing, which makes the terminal effectively unusable for me without reaching across the keyboard for the real Backspace key.

This patch makes Seance honour the remap.

Summary

xkb-level keysym remaps (e.g. KDE's Caps Lock behaves as Backspace) are silently dropped inside Seance. The remapped key sends nothing to the terminal even though wev confirms the Wayland compositor delivers the remapped keysym.

Root cause

GDK delivers a (keyval, hardware_keycode) pair where the keyval reflects the xkb remap but the hardware keycode still identifies the original physical key. handleKeyEvent in src/pane.zig forwards the raw keycode to ghostty_surface_key. libghostty's input handling is keycode-driven (ghostty/src/apprt/embedded.zig looks up the physical key in input/keycodes.zig by entry.native == keycode), so it sees the original physical key and ignores the remap.

For Caps Lock to Backspace on Linux: GDK reports keyval = BackSpace, keycode = 66 (Caps physical). libghostty sees keycode 66, identifies it as Caps Lock, and treats the event as a modifier press rather than producing Backspace input.

Fix

canonicalKeycode(keyval, keycode) returns the Linux evdev+8 keycode that natively produces the given keyval for a small set of named editing/navigation keys. handleKeyEvent uses the canonical keycode when it differs and strips GHOSTTY_MODS_CAPS in that case so the remap source's Lock state doesn't leak into the event.

Repro

1. KDE System Settings → Keyboard → Layouts → Advanced → "Caps Lock behaviour" → Make Caps Lock an additional Backspace (or any xkb option that remaps Caps to BackSpace).
2. In Seance: pressing Caps Lock does nothing; real Backspace works.
3. wev confirms sym: BackSpace, utf8: '\b' for the Caps press.
4. With this patch: Caps acts as Backspace inside Seance.

Notes / questions for review

• Scope. Limited to the named keys most likely to be remap targets (BackSpace, Tab, Return, KP_Enter, Escape, Delete, Insert, Home, End, PgUp/PgDn, arrows). I deliberately did not expand beyond what was needed to fix the reported case; happy to add F-keys etc. if you'd prefer.
• Hardcoded keycodes. gdk_display_map_keyval would query GDK for keycodes that produce a given keyval and avoid the hardcoded table. I went with the table because it allocates nothing on a hot input path, the editing-key set is small and stable on Linux, and Seance is already Linux-only per build.zig. Open to switching if you'd prefer.
• Alternative location. The same issue could be fixed inside libghostty (use keysym, not keycode, for non-modifier keys). The Seance-side fix is pragmatic; let me know if you'd rather see this upstream in your ghostty fork.
• Tested manually with a release build under KDE Plasma 6 / Wayland on Bazzite. No automated tests touched.

Test plan

• zig build succeeds
• Caps Lock → Backspace works in patched Seance under KDE Wayland (Colemak layout)
• Real Backspace still works
• No regressions observed in regular typing, arrow keys, or modifier-letter shortcuts

[raviboth]

This also resolves the reported case in #3 (filed for caps:swapescape). Same root cause: the raw hardware keycode is forwarded to libghostty instead of one reflecting the post-XKB keyval, so the embedded Ghostty resolves the physical key and ignores the remap.

This patch is essentially fix (a) proposed in #3 — for a small set of named editing/navigation keys, translate keyval to the canonical evdev+8 keycode and pass that to ghostty_input_key_s. Escape is in the table, so the caps-to-esc direction that #3 reports (physical CapsLock should produce Escape) now works. The reverse half of swapescape (physical Esc producing Caps_Lock) is a modifier and out of scope here.

The broader fix (b) in #3 — build an xkb_keymap from the GDK display and hand it to libghostty so Ghostty's own resolution honors the full session layout — is more correct but more invasive. Happy to take that route instead if you'd prefer it.

Note this is orthogonal to #6: that PR fixes the printable-text / IME path (text_ptr, consumed mods) and never touches .keycode, so it does not address keycode-driven remaps. Both edit handleKeyEvent, so whichever lands first I am glad to rebase on.

Two confirmed reporters now via XKB-level remaps (this and #3), so it is not a one-off config.