### System details

Intel Core 5 210H, Intel Raptor Lake-P + NVIDIA RTX 4050 Mobile, Omarchy 4.0.4-1, Hyprland 0.56.2

### What's wrong?

On a scrolling-layout workspace, a window that was put into tiled fullscreen
(`SUPER + CTRL + F`) and then taken through the ordinary fullscreen toggle
(`SUPER + F`) comes back at full width inside a half-width column, and one
`gaps_out` short on the left.

`fullscreenClient` itself is fine. Hyprland restores it to `2` on the way out
because that is the state the window came from, which is correct. What is not
restored is the column's width and origin.

**Repro** (scrolling layout, `scrolling.column_width = 0.5`, three columns on
the workspace):

| Step | Action | `fullscreen/fullscreenClient` | Geometry |
|---|---|---|---|
| 1 | — | `0/0` | `749x910 at 12` |
| 2 | `SUPER + CTRL + F` | `0/2` | `749x910 at 12` |
| 3 | `SUPER + F` | `2/2` | `1536x960 at 0` |
| 4 | `SUPER + F` | `0/2` | **`1527x910 at 2`** |
| 5 | `SUPER + CTRL + F` | `0/0` | `749x910 at 2` |

Steps 2 and 4 carry identical state fields and different geometry, which is
the whole bug: `0/2` means `749x910` at step 2 and `1527x910` at step 4.

Step 5 shows the layout can still size the window correctly — dropping the
client flag snaps the width back on its own.

`SUPER + ALT + F` ("Full width", `mode = "maximized"`) is affected the same
way; it returns `1507x910` instead of `749x910`.

**Expected:** leaving fullscreen returns the window to its column, `749x910 at
12`, with `fullscreenClient = 2` preserved.

**Actual:** `1527x910 at 2` — a full-width window inside a half-width column,
overlapping its neighbour, with the left gap visually gone.

**Notes**

- Both halves of the restore are wrong independently. Width is off by roughly
  a factor of two; the origin is off by exactly `gaps_out` (10px here).
- The origin error is self-healing: focusing another column, or any action
  that makes the layout recompute, puts it back. The width error is not.
- `omarchy-hyprland-window-tiled-fullscreen-toggle` itself is sound. Its
  primary dispatch works and it toggles cleanly when nothing else intervenes.

**Separate, smaller finding in the same script**

Its fallback line is dead on current Hyprland:

```
$ hyprctl dispatch fullscreenstate 0 0
error: [string "return hl.dispatch(fullscreenstate 0 0)"]:1: ')' expected near '0'
```

The pre-Lua dispatcher syntax no longer parses. It never fires while the
primary `hl.dsp.window.fullscreen_state` path works, but under `set -euo
pipefail` a failure of the primary path would exit 7 rather than fall through
to the fallback.

**Where the fix probably belongs**

The layout can still derive the correct geometry; the fullscreen-exit path just
never asks it to. Two separate nudges prove it, and they map onto the two
separate errors:

- Dropping `fullscreenClient` to `0` and raising it back to `2` restores the
  **width** (`1527 -> 749`), with no other change.
- A zero-width `colresize +0` restores the **origin** (`at 2 -> at 12`),
  touching neither the width nor the focus.

So this looks less like bad geometry maths and more like the exit path not
triggering a column recompute.

**Settings that matter**

```
Hyprland  0.56.2 (efb5099)
Monitor   eDP-1, 1920x1200 @ scale 1.25 (logical 1536x960), reserved top 26
Layout    scrolling (per-workspace rule), scrolling.column_width = 0.5
Gaps      gaps_out 10, gaps_in 5, border_size 2
```

Happy to attach `omarchy-debug` output if it would help, though this reproduces
from the settings above alone.