feat: add vLLM tuning pipeline + recompile voice pipelines with MLflow

New:
- vllm_tuning_pipeline.py: A/B benchmark different vLLM configs,
  logs latency/TPS/TTFT to MLflow (vllm-tuning experiment)
- vllm_tuning_pipeline.yaml: compiled KFP YAML

Updated:
- voice_pipeline.py: per-step NamedTuple outputs with latency tracking,
  new log_pipeline_metrics MLflow component
- voice_pipeline.yaml, tts_pipeline.yaml, rag_pipeline.yaml: recompiled

tts_pipeline.yaml

@@ -0,0 +1,87 @@
+# PIPELINE DEFINITION
+# Name: text-to-speech-pipeline
+# Description: Simple text to speech pipeline
+# Inputs:
+#    text: str
+components:
+  comp-synthesize-speech:
+    executorLabel: exec-synthesize-speech
+    inputDefinitions:
+      parameters:
+        text:
+          parameterType: STRING
+        tts_url:
+          defaultValue: http://ai-inference-serve-svc.ai-ml.svc.cluster.local:8000/tts
+          isOptional: true
+          parameterType: STRING
+    outputDefinitions:
+      parameters:
+        audio_b64:
+          parameterType: STRING
+        latency_s:
+          parameterType: NUMBER_DOUBLE
+deploymentSpec:
+  executors:
+    exec-synthesize-speech:
+      container:
+        args:
+        - --executor_input
+        - '{{$}}'
+        - --function_to_execute
+        - synthesize_speech
+        command:
+        - sh
+        - -c
+        - "\nif ! [ -x \"$(command -v pip)\" ]; then\n    python3 -m ensurepip ||\
+          \ python3 -m ensurepip --user || apt-get install python3-pip\nfi\n\nPIP_DISABLE_PIP_VERSION_CHECK=1\
+          \ python3 -m pip install --quiet --no-warn-script-location 'kfp==2.12.1'\
+          \ '--no-deps' 'typing-extensions>=3.7.4,<5; python_version<\"3.9\"'  &&\
+          \  python3 -m pip install --quiet --no-warn-script-location 'httpx' && \"\
+          $0\" \"$@\"\n"
+        - sh
+        - -ec
+        - 'program_path=$(mktemp -d)
+
+
+          printf "%s" "$0" > "$program_path/ephemeral_component.py"
+
+          _KFP_RUNTIME=true python3 -m kfp.dsl.executor_main                         --component_module_path                         "$program_path/ephemeral_component.py"                         "$@"
+
+          '
+        - "\nimport kfp\nfrom kfp import dsl\nfrom kfp.dsl import *\nfrom typing import\
+          \ *\n\ndef synthesize_speech(\n    text: str,\n    tts_url: str = \"http://ai-inference-serve-svc.ai-ml.svc.cluster.local:8000/tts\"\
+          \n) -> NamedTuple(\"TTSResult\", [(\"audio_b64\", str), (\"latency_s\",\
+          \ float)]):\n    \"\"\"Convert text to speech using TTS service.\"\"\"\n\
+          \    import base64\n    import time\n    import httpx\n    from collections\
+          \ import namedtuple\n\n    start = time.perf_counter()\n    with httpx.Client(timeout=120.0)\
+          \ as client:\n        response = client.post(\n            f\"{tts_url}/v1/audio/speech\"\
+          ,\n            json={\n                \"input\": text,\n              \
+          \  \"voice\": \"en_US-lessac-high\",\n                \"response_format\"\
+          : \"wav\"\n            }\n        )\n        audio_b64 = base64.b64encode(response.content).decode(\"\
+          utf-8\")\n    latency = time.perf_counter() - start\n\n    TTSResult = namedtuple(\"\
+          TTSResult\", [\"audio_b64\", \"latency_s\"])\n    return TTSResult(audio_b64,\
+          \ latency)\n\n"
+        image: python:3.13-slim
+pipelineInfo:
+  description: Simple text to speech pipeline
+  name: text-to-speech-pipeline
+root:
+  dag:
+    tasks:
+      synthesize-speech:
+        cachingOptions:
+          enableCache: true
+        componentRef:
+          name: comp-synthesize-speech
+        inputs:
+          parameters:
+            text:
+              componentInputParameter: text
+        taskInfo:
+          name: synthesize-speech
+  inputDefinitions:
+    parameters:
+      text:
+        parameterType: STRING
+schemaVersion: 2.1.0
+sdkVersion: kfp-2.12.1