Add Metal render pass state caching and sampler cache.

[Hparty]

Summary

• Add render pass state caching in MetalRenderPass to skip redundant Metal API calls for scissor rect, cull mode, uniform buffers, textures, samplers and vertex buffers when state has not changed.
• Add sampler cache in MetalGPU to reuse sampler state objects across frames.
• Extract MetalPixelFormatToPixelFormat from Backend.cpp to shared PixelFormatUtil, using hardcoded constants to avoid Metal header dependency.
• Add ShaderVisibility flags to BindingEntry, allowing precise per-stage uniform buffer binding on Metal instead of binding to both stages unconditionally.

Test plan

• Verify existing tests pass with no regressions
• Profile Metal rendering to confirm reduced API call overhead

[shlzxjp]

这个函数和MetalPixelFormatToPixelFormat的作用有什么不一样？看实现差不多