feat(wgpu_context): implement Windows DX12-first Vulkan-fallback policy

[GoCoder7]

Closes #99

Hi everyone,

1. Summary of Changes

This pull request implements an ordered GPU backend selection policy on Windows in wgpu_context: prioritizing DirectX 12 by default, providing automatic sequential fallback to Vulkan when DX12 is unavailable, and strictly respecting explicit user/environment overrides (WGPU_BACKEND).

Key Scope Points

1. Authoritative Context Seam (wgpu_context):
   • Updates DeviceHandle::new_from_compatible_surface_with_backends to execute sequential backend fallback.
   • Adds override_backends: Option<wgpu::Backends> and constructor with_features_limits_and_backends to WGPUContext.
   • Re-exports pub use wgpu;.
2. Zero Downstream Renderer Changes:
   • anyrender_vello_hybrid and anyrender_vello require no code changes. Both delegate surface and device initialization to DeviceHandle::new_from_compatible_surface, immediately inheriting the policy.
3. Downstream Consumers:
   • Embedders (such as Blitz or dioxus-native) require no first-wave code changes. Consumer-side validation and eventual removal of ad-hoc env hacks are rollout benefits, not part of the upstream patch scope.

---

2. Exact Seam Touched

Crate	File	Changes	Role
wgpu_context	crates/wgpu_context/src/lib.rs	+150 / -34	Ordered backend enumeration, compatibility check, and sequential device fallback loop.

---

3. Policy Semantics

The selection logic follows a strict 3-tier sequence:

[Window / Device Handle Request]
              │
              ▼
   Is WGPU_BACKEND set in env,
   or was backends explicitly passed?
         ├── YES ──► Use specified backend directly (Honors override)
         │
         └── NO  ──► Is target_os = "windows"?
                         ├── NO  ──► Default wgpu adapter resolution
                         │
                         └── YES ──► Tier 1: Try DirectX 12
                                         │
                                         ├─ Found compatible adapter & device created ──► Return DX12 Device
                                         │
                                         └─ Failed or no adapter ──► Tier 2: Try Vulkan Fallback
                                                                         │
                                                                         ├─ Found compatible adapter & device created ──► Return Vulkan Device
                                                                         │
                                                                         └─ Failed ──► Tier 3: Exhaustive default fallback

---

4. Validation Evidence & Proof Boundaries

Testing was performed on Windows 10 x64 hardware with Intel(R) HD Graphics 520 (Skylake generation, driver 27.20.100.8854):

1. Compilation:
   • cargo check -p wgpu_context: Passed cleanly with exit code 0.
   • Downstream integration check (browser / dioxus-native consuming this patch): Compiled cleanly without downstream code changes.
2. Default Execution (WGPU_BACKEND unset):
   • Executed downstream window with WGPU_BACKEND unset.
   • Verified active adapter: adapter: name='Intel(R) HD Graphics 520', backend=Dx12, driver='27.20.100.8854'.
   • DXGI swapchain presented cleanly to DWM; the blank white screen bug caused by Intel Windows Vulkan swapchain failure was completely resolved.
3. Explicit Override Sanity:
   • Executed with $env:WGPU_BACKEND = "vulkan".
   • Verified active adapter: backend=Vulkan. Confirmed that explicit user/CI intent is never overridden.
4. Proof Scope (Honesty Note):
   • The default DX12 path and explicit override semantics are directly runtime-proven.
   • The automatic DX12 failure -> Vulkan fallback branch is implementation-backed and verified by code-path review, but was not directly observed on hardware where DX12 itself fails.

Thanks!