fix: use test.pypi for uv sync on release branches

The previous approach tried to install before uv sync, but there's no
venv yet. The correct solution:

- Release branches: Point UV_INDEX_URL to test.pypi so uv sync can
  resolve RC versions, then install exact git version after sync
- Non-release branches: Run uv sync normally, then install git version
  if client-version=latest

This lets uv sync create the venv first, then we install/override the
client version as needed.

.github/actions/setup-runner/action.yml

@@ -18,13 +18,17 @@ runs:
         python-version: ${{ inputs.python-version }}
         version: 0.7.6
 
-    - name: Install llama-stack-client
+    - name: Configure client installation
+      id: client-config
       uses: ./.github/actions/install-llama-stack-client
       with:
         client-version: ${{ inputs.client-version }}
 
     - name: Install dependencies
       shell: bash
+      env:
+        UV_INDEX_URL: ${{ steps.client-config.outputs.uv-index-url }}
+        UV_EXTRA_INDEX_URL: ${{ steps.client-config.outputs.uv-extra-index-url }}
       run: |
         echo "Updating project dependencies via uv sync"
         uv sync --all-groups
@@ -32,5 +36,11 @@ runs:
         echo "Installing ad-hoc dependencies"
         uv pip install faiss-cpu
 
+        # Install specific client version after sync if needed
+        if [ "${{ steps.client-config.outputs.install-after-sync }}" = "true" ]; then
+          echo "Installing llama-stack-client from: ${{ steps.client-config.outputs.install-source }}"
+          uv pip install ${{ steps.client-config.outputs.install-source }}
+        fi
+
         echo "Installed llama packages"
         uv pip list | grep llama