fix(sokol): remove HiDPI camera offset double-scaling

[apotema]

Problem

On HiDPI mobile (Android, 2400×1080 physical / 1024×768 design), world-space entities render far off-screen. The flying-platform scene appears empty because the ship carcase ends up at NDC (1.34, ~−2.4) — completely outside the visible range.

Root cause

a610afd added cam.offset * (screen / design) scaling in toNdcX/toNdcY/screenToWorld/worldToScreen under the assumption that the camera offset was in design coordinates. It isn't.

labelle-gfx's camera.toBackend() builds the offset as:

```zig
pub fn toBackend(self: *const Self) BackendImpl.Camera2D {
const dims = self.getViewportDimensions();
return .{
.offset = .{ .x = dims.width / 2.0, .y = dims.height / 2.0 },
...
};
}

pub fn getViewportDimensions(self: *const Self) struct { width: f32, height: f32 } {
...
return .{
.width = @floatFromInt(BackendImpl.getScreenWidth()),
.height = @floatFromInt(BackendImpl.getScreenHeight()),
};
}
```

In the sokol backend getScreenWidth/getScreenHeight return screen_w/screen_h — the physical framebuffer dimensions set by setScreenSize(). So by the time beginMode2D(camera) stores it, cam.offset is already in physical pixels.

Multiplying that already-physical value by screen/design (≈ 2.3 on this device) shoves the world origin off-screen:

```
cam.offset.x = screen_w / 2 = 1200
1200 * (2400 / 1024) = 2812.5
NDC = (2812.5 / 2400) * 2 - 1 = 1.34 // outside [-1, 1]
```

Fix

Drop the * (fw/fdw) / * (fh/fdh) correction from all four functions and use cam.offset directly. Dividing physical-pixel screen_x by physical screen_w already yields correct NDC: (1200 / 2400) * 2 - 1 = 0.

The screen-space (no-camera) branch still divides by design_w/design_h — screen-space positions are still authored in design coordinates, so that path is untouched.

Test plan

• Deploy to flying-platform-labelle Android emulator build
• Verify world-space entities render at expected positions (ship carcase centered in view, not offset off-screen)
• Screen-space background still fills the whole screen
• Desktop sokol path unchanged (design == physical → ratio = 1, so the old math collapsed to the same thing; still want a sanity check)

[cursor]

PR Summary

Medium Risk
Changes core camera coordinate transforms in the Sokol backend; low blast radius outside rendering, but mistakes would visibly break world/screen mapping on HiDPI or non-HiDPI devices.

Overview
Fixes HiDPI camera positioning in the Sokol backend by removing the extra screen/design scaling applied to Camera2D.offset.

toNdcX/toNdcY, screenToWorld, and worldToScreen now use camera.offset directly (treated as physical pixels), while the screen-space (no-camera) NDC mapping continues to use design_w/design_h. The camera math comments were updated to document this contract.

^{Reviewed by Cursor Bugbot for commit df8260e. Bugbot is set up for automated code reviews on this repo. Configure here.}

[gemini-code-assist]

Code Review

This pull request simplifies coordinate transformations in the Sokol backend by removing redundant HiDPI scaling for camera offsets. The logic was updated in toNdcX, toNdcY, screenToWorld, and worldToScreen because the camera offset is already provided in physical pixels, making the manual scaling between design and physical dimensions unnecessary. I have no feedback to provide.

[Copilot]

Pull request overview

Fixes incorrect camera coordinate transforms in the sokol backend on HiDPI/mobile by removing an extra (and incorrect) design→physical scaling of Camera2D.offset, which was pushing world-space content off-screen when screen != design.

Changes:

• Remove double-scaling of cam.offset in toNdcX / toNdcY when a camera is active.
• Remove the same offset scaling from screenToWorld and worldToScreen.
• Update the in-file documentation to clarify that cam.offset is already in physical framebuffer pixels for the sokol backend.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.