Update requirements/cuda.txt to clarify the pinning of nvidia-cublas-cu12 version 12.9.1.4 due to dependency resolution issues with torch. This comment aids in understanding the force-reinstall strategy in Dockerfile.ci.

Signed-off-by: Taichang Zhou <tzhouam@connect.ust.hk>

requirements/cuda.txt

@@ -1,4 +1,5 @@
 -r common.txt
 onnxruntime>=1.23.2
 fa3-fwd==0.0.2
-nvidia-cublas-cu12==12.9.1.4
+# nvidia-cublas-cu12 is pinned to 12.9.1.4 via force-reinstall in Dockerfile.ci
+# because torch pins its own cublas version, making it unsatisfiable at resolve time.