Add result layer wrappers for 2D/3D pitched memory operations

[wizenink]

[DRAFT] Adds low-level support for 2D/3D pitched memory allocation and transfers. Safe API wrappers
to follow.

Changes

• Result layer: Added wrappers for pitched memory operations:
  • malloc_pitched() - Allocate 2D memory with GPU-optimized row alignment
  • memcpy_2d_async() / memcpy_2d_sync() - 2D memory transfers
  • memcpy_3d_async() / memcpy_3d_sync() - 3D memory transfers
• Example: Added 11-pitched-memory.rs demonstrating:
  • Pitched allocation (512-byte aligned rows for 32-byte data)
  • Host <-> Device 2D transfers
  • Sub-region (strided) copying
  • Data verification

Usage (unsafe for now)

let (device_ptr, pitch) = unsafe {
    result::malloc_pitched(width_bytes, height, elem_size)?
};

let copy_params = sys::CUDA_MEMCPY2D { /* ... */ };
unsafe { result::memcpy_2d_async(&copy_params, stream)? };

TODO (before marking ready for review)

• Design safe API (CudaPitched2D, CudaPitched3D types)
• Implement builder pattern for copy operations
• Add safe wrapper methods to CudaStream

[wizenink]

@coreylowman Any tips or preferences on how to design the safe API? Might have some time to implement this later tonight.
BTW, thank you for your work, really loving this bindings, very idiomatic :)