Manual verification needed: hot-reload swap actually updates a mesh in the running editor

[JeanPhilippeKernel]

Summary

PR #748 implements and code/test-verifies the RenderResourceManager hot-reload swap mechanism (ScheduleSwap/FlushPendingSwaps) — editing and re-saving a mesh asset while the editor is open should now update the live GPU buffer instead of silently doing nothing (the previous behavior, see #740).

What's missing is an actual manual, visual, end-to-end check in the running editor: import a mesh, place it in a scene, edit the source file on disk, and confirm the mesh visibly updates in the viewport without a restart.

Why this wasn't done as part of #748

1. SampleProject's DefaultScene has no mesh assets loaded at all today (just a DirectionalLight) — there's nothing already in the scene to hot-reload against.
2. Getting one in requires importing a mesh through the editor GUI (drag-and-drop / import dialog), which isn't automatable in this environment:
   • cliclick's synthetic mouse clicks don't reach the Obelisk window (missing Accessibility/Input Monitoring permission for the driving process).
   • AppleScript's System Events can move/resize the window but can't click inside its custom-rendered ZUI content (no native accessibility elements exposed).

What's needed to close this

Either:

• Manual pass: import a test mesh (e.g. a simple glTF/OBJ) into SampleProject, place it in DefaultScene, run the editor, edit the source mesh file (change a vertex, re-export), save, and confirm the in-viewport mesh updates within a few frames without restarting. Also worth checking that the previous geometry doesn't flash/corrupt during the swap.
• Or: grant Accessibility/Input Monitoring permission to whatever process drives future automated GUI testing, so this (and similar future UI-driven checks) can be scripted.

Where

• Mechanism under test: ZEngine/ZEngine/Rendering/RenderResourceManager.cpp (ScheduleSwap, FlushPendingSwaps, AppendMeshData)
• Related: RenderResourceManager: SwapKind::Buffer silently dropped on hot-reload; no ABA protection on handle pool #740 (original bug), PR Rendering foundation hardening: real RenderGraph sort, GpuAllocator pools, RRM hot-reload #748 (the fix)

[JeanPhilippeKernel]

Manually verified, twice, controlling for the engine log's ~10-15s flush lag to rule out a false negative:

1. Imported a hand-written test cube (1x scale) into DefaultScene, confirmed it uploaded and rendered at baseline size from a fixed camera position.
2. Edited the source to 3x scale, re-imported the same file (same MeshUUID, preserved via .meta).
3. Waited 20+ seconds — no new [RRM] log line at all (no upload, no swap attempt, no failure).
4. Screenshotted from the identical camera position — cube unchanged, still 1x.

Result: hot-reload does not update the live mesh. Root-caused to AssetManager::IngestMesh silently early-returning whenever a MeshUUID is already registered (AssetManager.cpp:106-107) — before ever reaching the code that would mark the asset stale and let RenderResourceManager::ScheduleSwap (from #748, implemented correctly) do anything. Both the manual re-import path and the file-watcher hot-reload path share this same dead end.

Filed as #762 with full repro + root cause. Closing this one since the manual verification it asked for is done — the answer is "it doesn't work," and #762 tracks the actual fix.