feat: add GCP Vertex.AI hyperparameter tuning handler

[Aanushka001]

Description

This PR adds a new event handler VertexAILogger that enables PyTorch Ignite training code to report metrics to GCP Vertex AI's hyperparameter tuning service.

Fixes #3023

What's included:

• VertexAILogger - A BaseLogger subclass that writes metrics in the format expected by Vertex AI's hyperparameter tuning service. It reads the CLOUD_ML_HP_METRIC_FILE environment variable (default: /tmp/hypertune/output.metrics) and writes newline-delimited JSON metric entries, following the same protocol as the official cloudml-hypertune library.
• OutputHandler - Helper handler to log engine output/metrics via report_hyperparameter_tuning_metric().
• OptimizerParamsHandler - Helper handler to log optimizer parameters (e.g., learning rate).

Usage example

from ignite.handlers.gcp_vertex_logger import VertexAILogger
from ignite.engine import Events

vertex_logger = VertexAILogger()

# Report accuracy after each epoch
vertex_logger.attach_output_handler(
    trainer,
    event_name=Events.EPOCH_COMPLETED,
    tag="training",
    metric_names=["accuracy"],
    global_step_transform=lambda engine, event_name: engine.state.epoch,
)

# Report at the end of training
vertex_logger.attach_output_handler(
    trainer,
    event_name=Events.COMPLETED,
    tag="training",
    metric_names=["accuracy"],
    global_step_transform=lambda engine, event_name: engine.state.epoch,
)

How it works

Vertex AI hyperparameter tuning works by reading metric values from a file at the path specified by the CLOUD_ML_HP_METRIC_FILE environment variable. This handler writes the metrics in the correct newline-delimited JSON format, allowing Vertex AI to evaluate trial performance. No external dependencies are required.

Checklist

• Handler follows the existing BaseLogger pattern (like WandBLogger, MLflowLogger)
• Exported via ignite/handlers/__init__.py
• Full type hints and docstrings
• No new external dependencies