Implement Vulkan fast path for NfRectMaskContent

[elcritch]

Summary

NfRectMaskContent now has fast-path implementations for Metal and OpenGL. Vulkan should get the same efficient rounded-rect mask behavior instead of relying on the regular clip/mask fallback.

Current behavior

For compatibility, Vulkan should treat NfRectMaskContent as a regular clip for now:

• beginRectMask(maskRect, radii) should behave like beginMask(maskRect, radii) followed by endMask().
• popRectMask() should behave like popMask().

That preserves rendering correctness while the fast path is unimplemented.

Desired implementation

Add a Vulkan fast path comparable to the Metal/OpenGL implementations:

• avoid flushing/drawing a mask queue for leaf GUI items that only need rounded-rect clipping behavior;
• carry rect-mask params/radii/transform data through the Vulkan vertex path only for batches that actually contain rect masks;
• keep normal non-clip batches on the existing path so regular scenes do not pay per-vertex or shader cost;
• support mixed batches: unmasked sibling, rect-masked sibling, then unmasked sibling again;
• fall back to regular clipping for nested rect masks or unsupported cases.

Suggested tests/benchmarks

• Reuse the mixed-batch regression render test in tests/trender_layers_clip.nim with Vulkan enabled.
• Compare examples/windy_clip_mask_benchmark.nim for clip + sub-clip vs clip + rect-mask on Vulkan.
• Run examples/windy_non_clip_benchmark.nim to confirm normal non-clip scenes are not taxed.

Notes

Metal and OpenGL currently have fast paths. Vulkan should remain correct via regular clipping until this issue is implemented.