## Description

The Omarchy shell (`quickshell`) leaks VRAM on every failed layer-shell surface creation, instead of releasing the partial allocation. On a system with limited VRAM, this creates a self-reinforcing failure loop: once GPU memory is tight enough that one surface creation fails, every subsequent attempt (e.g. repeatedly pressing Super+Space to open the menu) leaks more VRAM and makes recovery less likely, until `omarchy restart shell` is run to reclaim it.

## Steps to reproduce

1. Run Omarchy on a machine with a small-VRAM discrete GPU shared across several GPU-memory-hungry apps (in my case: an NVIDIA GTX 1050 with 2GB VRAM, alongside Chromium, VS Code, and other Electron apps).
2. Let GPU memory usage climb until there isn't enough free VRAM for `quickshell` to allocate a new EGL/GL context (I hit this around 1729/2048 MiB used, ~85%).
3. Press `SUPER + SPACE` (Omarchy menu) — nothing appears.
4. Keep pressing it — the shell keeps trying, keeps failing, and its own VRAM footprint keeps growing with each failed attempt (in my case: 392 MiB right after a clean shell restart, growing to 702 MiB, then 811 MiB within ~2 minutes, without a single successful render).

## Expected behavior

- A failed layer-shell surface creation should not leak the partially-allocated GPU resources.
- Ideally, quickshell would detect repeated `EGL_BAD_ALLOC` / "Failed to create graphics context" failures and either back off, or the Omarchy supervisor would notice sustained failure and auto-restart the shell (similar to the recovery gap noted in #8491, though that issue is about a different trigger — EGL context loss on resume — this one is triggered purely by VRAM exhaustion, no suspend/resume involved).

## Actual behavior

`SUPER + SPACE` (and any other menu/panel toggle) silently does nothing. The keybinding and IPC call both succeed — `hyprctl binds` shows the bind is correctly registered, and `omarchy-menu toggle root` returns exit 0 — but no surface is ever rendered. journalctl shows repeated:

```
WARN: QWaylandGLContext::makeCurrent: eglError: 0x3003
WARN: QRhiGles2: Failed to make context current. Expect bad things to happen.
WARN: Failed to create RHI (backend 2)
ERROR: Failed to create graphics context for qs::wayland::layershell::WlrLayershell(...): "Failed to initialize graphics backend for OpenGL."
```

Only `omarchy restart shell` recovers it (until VRAM fills up again).

## System details

- Omarchy: 4.0.3-1
- Hyprland: 0.56.2
- Quickshell: 0.3.1
- Kernel: 7.2.3-arch1-3
- GPU: NVIDIA GeForce GTX 1050, 2048 MiB VRAM, driver 580.178.04

## Notes

Possibly related to #8491 (shell stops rendering after resume-time `EGL_BAD_CONTEXT` on NVIDIA) — same failure class (Qt RHI/EGL context loss on NVIDIA with no automatic recovery), but a different trigger (VRAM exhaustion under normal use vs. suspend/resume).