perf: parallelize model downloads and switch to registry cache

- Parallelize all 14 model downloads using ThreadPoolExecutor (6 workers)
  Downloads were sequential (~30 min), now concurrent (~5-10 min)
- Switch Docker cache from type=gha to type=registry (GHCR)
  GHA cache has 10 GB limit causing blob eviction and corrupted builds
  Registry cache has no size limit and persists across runner instances
- Add pip download cache mounts to all pip install layers
  Prevents re-downloading packages when layers rebuild

.github/workflows/ci.yml

@@ -99,6 +99,13 @@ jobs:
 
       - uses: docker/setup-buildx-action@v3
 
+      - name: Log in to GHCR (for registry cache)
+        uses: docker/login-action@v3
+        with:
+          registry: ghcr.io
+          username: ${{ github.repository_owner }}
+          password: ${{ secrets.GHCR_TOKEN }}
+
       - uses: docker/build-push-action@v6
         with:
           context: .
@@ -106,5 +113,5 @@ jobs:
           push: false
           tags: ashim:ci
           build-args: SKIP_MODEL_DOWNLOADS=true
-          cache-from: type=gha,scope=unified
-          cache-to: type=gha,mode=max,scope=unified
+          cache-from: type=registry,ref=ghcr.io/${{ github.repository }}:cache-linux-amd64
+          cache-to: type=registry,ref=ghcr.io/${{ github.repository }}:cache-ci,mode=max

.github/workflows/release.yml

@@ -121,8 +121,8 @@ jobs:
           platforms: ${{ matrix.platform }}
           labels: ${{ steps.meta.outputs.labels }}
           outputs: type=image,"name=ashimhq/ashim,ghcr.io/${{ github.repository }}",push-by-digest=true,name-canonical=true,push=true
-          cache-from: type=gha,scope=${{ env.PLATFORM_PAIR }}
-          cache-to: type=gha,mode=max,scope=${{ env.PLATFORM_PAIR }}
+          cache-from: type=registry,ref=ghcr.io/${{ github.repository }}:cache-${{ env.PLATFORM_PAIR }}
+          cache-to: type=registry,ref=ghcr.io/${{ github.repository }}:cache-${{ env.PLATFORM_PAIR }},mode=max
 
       - name: Export digest
         run: |