feat: simplify CI to single unified Docker build

Remove 3-variant matrix (full/lite/cuda). Single build produces
a multi-arch manifest (amd64 + arm64) pushed to Docker Hub and GHCR.
Tags: latest, X.Y.Z, X.Y, X. CI builds native platform only (amd64)
for speed. Multi-arch only on release.

.github/workflows/ci.yml

@@ -80,20 +80,8 @@ jobs:
       - run: pnpm build
 
   docker:
-    name: Docker Build Test (${{ matrix.tag }})
+    name: Docker Build Test
     runs-on: ubuntu-latest
-    strategy:
-      matrix:
-        include:
-          - tag: full
-            variant: full
-            gpu: "false"
-          - tag: lite
-            variant: lite
-            gpu: "false"
-          - tag: cuda
-            variant: full
-            gpu: "true"
     steps:
       - uses: actions/checkout@v4
 
@@ -104,9 +92,6 @@ jobs:
           context: .
           file: docker/Dockerfile
           push: false
-          build-args: |
-            VARIANT=${{ matrix.variant }}
-            GPU=${{ matrix.gpu }}
-          tags: stirling-image:ci-${{ matrix.tag }}
-          cache-from: type=gha,scope=${{ matrix.tag }}
-          cache-to: type=gha,mode=max,scope=${{ matrix.tag }}
+          tags: stirling-image:ci
+          cache-from: type=gha,scope=unified
+          cache-to: type=gha,mode=max,scope=unified