llama.cpp b11013 fixes Vulkan buffer_reference alignment in im2col shaders

The llama.cpp b11013 release addresses a validation error in Vulkan by correcting buffer_reference alignment in im2col and im2col_3d shaders. The fix resolves issue #28960 by aligning the pointer stride to actual write sizes, eliminating GPU-AV validation hits.

• Fixes VUID-RuntimeSpirv-PhysicalStorageBuffer64-06315 triggered by misaligned writes in im2col.comp and im2col_3d.comp.
• Sets buffer_reference_align = D_SIZE to match the per-variant define for float (4) and float16_t (2).
• Reduces validation hits from 20 to 0 for both IM2COL and IM2COL_3D operations.

This change ensures correct Vulkan validation behavior for GPU-accelerated inference on supported platforms.