MoE prefill bf16 perf improvement for qwen-3.5-35B-A3B (#18829)

|                    | Baseline  | Batched    | Speedup |
|--------------------|-----------|------------|---------|
| Prefill (1341 tok) | 588 tok/s | 1807 tok/s | 3.07x   |
| Decode (128 tok)   | 90 tok/s  | 86 tok/s   | ~1.0x   | (noise?)

Author: digantdesai

backends/cuda/CMakeLists.txt

@@ -109,7 +109,9 @@ set(_aoti_cuda_shim_sources runtime/shims/memory.cpp
 
 # Only build int4mm shim when CUDA language/toolchain is available.
 if(CMAKE_CUDA_COMPILER)
-  list(APPEND _aoti_cuda_shim_sources runtime/shims/int4mm.cu)
+  list(APPEND _aoti_cuda_shim_sources runtime/shims/int4mm.cu
+       runtime/shims/sort.cu
+  )
 endif()
 
 add_library(aoti_cuda_shims SHARED ${_aoti_cuda_shim_sources})