fix: restrict SGLang CUDA build to linux/amd64 only

B200/H200 GPUs are x86_64 architecture only, and we install CUDA toolkit
from the x86_64 repository. Building for arm64 would fail.

ARM64 CUDA support can be added by the community if needed.

Author: dougyster

.github/workflows/release.yml

@@ -171,7 +171,7 @@ jobs:
         with:
           file: Dockerfile
           target: final-sglang
-          platforms: linux/amd64, linux/arm64
+          platforms: linux/amd64
           build-args: |
             "LLAMA_SERVER_VERSION=${{ inputs.llamaServerVersion }}"
             "LLAMA_SERVER_VARIANT=cuda"