[SOLID] [M21] Inject font atlas; mutex-guard graphics_presets

[MichaelFisher1997]

Part of #776 — SOLID architecture debt umbrella.
Phase: P3 · Finding: M21

Problem

Two mutable globals violate DIP:

• modules/engine-ui/src/font.zig:8 var active_atlas — hidden global pointer, no synchronization, set by UISystemManager and read on every drawText/measureTextWidth.
• modules/game-core/src/settings/json_presets.zig:51 pub var graphics_presets — process-wide mutable list, no mutex. Loaded once at startup, read from game-ui settings screens.

Evidence

(Full evidence and file:line citations are in the umbrella #776.)

Fix

1. Font atlas: pass it via injection through the UI draw context instead of a global.
2. graphics_presets: gate behind an accessor that holds a mutex, or make immutable after load.

Verification

• nix develop --command zig build test (includes shader validation)
• If rendering-touching: capture headless screenshot baseline before/after (-Dskip-present)
• If behavior-preserving refactor: golden-output test must stay green

Constraints

• Small reviewable PR targeting dev; conventional commit (refactor: / feat:)
• Preserve current behavior unless this issue explicitly changes it
• Keep worker-thread RHI isolation intact (no RHI calls off main thread)

Notes

Currently safe only because UI runs single-threaded on the main thread after single-shot init — an undocumented invariant. Fix removes the latent hazard.

Tracking: #776

[github-actions]

I have validated the file:line citations in the issue (modules/engine-ui/src/font.zig:8 and modules/game-core/src/settings/json_presets.zig:52) and traced all consumers. Posting a focused analysis comment.

## Analysis

Validated both citations and traced consumers. Issue is clear and well-scoped — classification: **Refactor** (DIP cleanup, not a runtime bug).

### Consumers traced

**`active_atlas` global** (5 references in `modules/engine-ui/src/`, 1 in `src/engine/ui/font.zig:10` re-export):
- Written: `ui_system_manager.zig:41` (deinit) and `ui_system_manager.zig:65` (every frame in `draw`)
- Read: `font.zig:43` and `font.zig:57` (`drawText` / `measureTextWidth`)
- Production call surface: `Font.drawText`, `Font.measureTextWidth`, `Font.drawTextCentered`, `Font.drawTextCenteredFit`, `Font.drawNumber` — all already take `u: *UISystem`, so the context already exists.

**`graphics_presets` global** (read across 4 files outside `json_presets.zig`):
- `src/game/app.zig:518`, `modules/game-ui/src/screens/graphics.zig:119-131`, `modules/game-ui/src/screens/settings.zig:377-388`, `modules/game-ui/src/settings_ui.zig:96-97`, `modules/game-core/src/settings/ui_helpers.zig:98-99`, plus `modules/game-core/src/settings/tests.zig:53`.
- All readers treat it as read-only after init — `initPresets`/`deinitPresets` write; everything else reads `items.len` / `items[i]`. No concurrent writers today, but the `pub var` shape allows them.

### Suggested implementation

**1. Font atlas injection** — store the optional atlas inside `UISystem` rather than in module-level state. `UISystem` is already the per-frame context that gets threaded through every draw call:
- Add `font_atlas: ?*const FontAtlas = null` to `UISystem` in `modules/engine-ui/src/ui_system.zig:12`.
- Add `pub fn setFontAtlas(self: *UISystem, atlas: ?*const FontAtlas) void` on `UISystem`.
- Replace `Font.active_atlas` reads with `u.font_atlas` in `font.zig:43` and `font.zig:57`.
- In `ui_system_manager.zig:65`, call `u.setFontAtlas(self.font_atlas)` (binding to the local `u` already unwrapped on line 62).
- Delete `var active_atlas`, `setActiveAtlas`, and the `src/engine/ui/font.zig:10` re-export. Search confirms no other external callers.
- `measureTextWidth` currently has no `UISystem` parameter — you will need to add `u: *UISystem` (or take the atlas as a separate parameter) and update callers in `font.zig:66, 71, 105`. Verify the ripgrep-cleanup: `rg "measureTextWidth"` to catch them all.

**2. `graphics_presets` access** — replace the `pub var` with a private field plus thread-safe / immutable accessors:
- Make `graphics_presets` module-private and wrap read operations (`getCount`, `getName`, `getIndex`, `apply`, internal list views) behind a `std.Thread.Mutex` inside the file. Since reads currently dominate and writes only happen in `initPresets`/`deinitPresets`, an RwLock or RCU-style fence works, but a plain Mutex matches the issue's stated approach and keeps the diff small.
- For tests (`modules/game-core/src/settings/tests.zig:53`), add a minimal `pub fn count() usize` accessor so they don't need to touch the mutex.
- Behavior is preserved because all current callers already operate on the list from the main thread — the mutex is a latent-hazard defense, not a behavior change, matching the "behavior-preserving refactor" constraint.

### Scope recommendation

The two fixes are independent (different modules, different call graphs). Recommend shipping as **two PRs** for cleaner review and easier revert — both targeting `dev`:
- `refactor(engine-ui): inject font atlas through UISystem (M21)`
- `refactor(game-core): mutex-guard graphics_presets accessor (M21)`

If combined into a single PR, keep it tight and split commit boundaries.

### Verification checklist

- `nix develop --command zig build test` (covers shader validation + unit tests; `modules/game-core/src/settings/tests.zig` exercises `getIndex`/`graphics_presets.items.len`)
- `nix develop --command zig fmt src/` per AGENTS.md
- Headless screenshot baseline (`-Dskip-present`) before/after, since the font path renders every frame
- Confirm no `measureTextWidth` callers are missed by grepping the repo
- Read-only test: `modules/game-core/src/settings/tests.zig:53` must stay green (golden-output test referenced in issue)

### No missing info

Reproduction is unnecessary — this is a structural refactor with no runtime symptom. The cited `file:line` references match the current `dev` HEAD. Proceeding to implementation does not require a stack trace, repro, or version.

opencode session  |  github run