Add tpu-info-streaming-verification DAG to validate streaming mode refresh frequency

[chengpinglin]

Description

This PR introduces a new automated DAG designed to verify the tpu-info CLI tool's --streaming functionality on TPU v6e-16 slices. The primary goal is to ensure that the tool correctly honors the requested refresh rate by validating the precise timing of hardware telemetry updates.

Key Changes:

• Metric Accuracy via $\Delta t$ Calculation: Implemented TPUPerformanceAnalyzer to parse microsecond-precision timestamps and validate the refresh rate based on the precise delta ($\Delta t$) between consecutive, unique hardware data frames.
• Tolerance & Scheduling Jitters: Implemented a ±20% tolerance buffer ($0.8 \times \text{Rate} \le \Delta t \le 1.2 \times \text{Rate}$). This accounts for system scheduling jitters—unavoidable variations in execution timing caused by OS kernel task scheduling and Kubernetes resource contention that can shift the metric capture window.
• Eager Refresh Strategy & Low-Latency Output: The logic recognizes the hardware driver's Eager Refresh Strategy, where the machine outputs telemetry prematurely (ahead of the exact interval) to prioritize data freshness.

Tests

• GCP Composer name: tony-test (under GCP project: cloud-ml-auto-solutions)
• GCP Composer version: 2.13.1

Checklist

Before submitting this PR, please make sure (put X in square brackets):

• I have performed a self-review of my code.
• I have necessary comments in my code, particularly in hard-to-understand areas.
• I have run one-shot tests and provided workload links above if applicable.
• I have made or will make corresponding changes to the doc if needed.