feat(ci): keep only one re-recording job because independent recordings will conflict (#2956)

A couple of important updates:

- When recording tests, we cannot be generating a matrix because all the
independent recordings will conflict.
- In fact, we just don't need a matrix on test types any more because
things are very fast and the overhead of `llama stack build` and setting
up `uv` etc. is much more.
- Refactored the running of tests into an independent action

.github/actions/run-integration-tests/action.yml

@@ -0,0 +1,73 @@
+name: 'Run Integration Tests'
+description: 'Run integration tests with configurable execution mode and provider settings'
+
+inputs:
+  test-types:
+    description: 'Test types to run (JSON array)'
+    required: true
+  stack-config:
+    description: 'Stack configuration: "ci-tests" or "server:ci-tests"'
+    required: true
+  provider:
+    description: 'Provider to use: "ollama" or "vllm"'
+    required: true
+  inference-mode:
+    description: 'Inference mode: "record" or "replay"'
+    required: true
+
+outputs:
+  logs-path:
+    description: 'Path to generated log files'
+    value: '*.log'
+
+runs:
+  using: 'composite'
+  steps:
+    - name: Run Integration Tests
+      env:
+        LLAMA_STACK_CLIENT_TIMEOUT: "300"
+        LLAMA_STACK_TEST_RECORDING_DIR: "tests/integration/recordings"
+        LLAMA_STACK_TEST_INFERENCE_MODE: ${{ inputs.inference-mode }}
+      shell: bash
+      run: |
+        stack_config="${{ inputs.stack-config }}"
+        EXCLUDE_TESTS="builtin_tool or safety_with_image or code_interpreter or test_rag"
+
+        # Configure provider-specific settings
+        if [ "${{ inputs.provider }}" == "ollama" ]; then
+          export OLLAMA_URL="http://0.0.0.0:11434"
+          export TEXT_MODEL="ollama/llama3.2:3b-instruct-fp16"
+          export SAFETY_MODEL="ollama/llama-guard3:1b"
+          EXTRA_PARAMS="--safety-shield=llama-guard"
+        else
+          export VLLM_URL="http://localhost:8000/v1"
+          export TEXT_MODEL="vllm/meta-llama/Llama-3.2-1B-Instruct"
+          EXTRA_PARAMS=""
+          EXCLUDE_TESTS="${EXCLUDE_TESTS} or test_inference_store_tool_calls"
+        fi
+
+        TEST_TYPES='${{ inputs.test-types }}'
+        echo "Test types to run: $TEST_TYPES"
+
+        for test_type in $(echo "$TEST_TYPES" | jq -r '.[]'); do
+          # if provider is vllm, exclude the following tests: (safety, post_training, tool_runtime)
+          if [ "${{ inputs.provider }}" == "vllm" ]; then
+            if [ "$test_type" == "safety" ] || [ "$test_type" == "post_training" ] || [ "$test_type" == "tool_runtime" ]; then
+              continue
+            fi
+          fi
+
+          echo "=== Running tests for: $test_type ==="
+
+          if uv run pytest -s -v tests/integration/$test_type --stack-config=${stack_config} \
+            -k "not( ${EXCLUDE_TESTS} )" \
+            --text-model=$TEXT_MODEL \
+            --embedding-model=sentence-transformers/all-MiniLM-L6-v2 \
+            --color=yes ${EXTRA_PARAMS} \
+            --capture=tee-sys | tee pytest-${{ inputs.inference-mode }}-$test_type.log; then
+            echo "✅ Tests completed for $test_type"
+          else
+            echo "❌ Tests failed for $test_type"
+            exit 1
+          fi
+        done