docs: add training and validation guides

[jpata]

Part of #500.

Summary

• add progressive guides for standard training, checkpoint continuation, fine-tuning, multi-GPU execution, Ray Train, and Ray Tune
• document four validation layers: dataset integrity, model behavior, physics behavior, and deployment behavior
• add focused dataset, Key4HEP, CMS, and ONNX validation workflows
• add an artifact and directory reference for reproducible model bundles
• extend the published CLD download example to all three standard configuration-1 datasets: ttbar, WW, and qq
• connect the new pages from the landing page, workflow chooser, dataset guides, and README
• use affirmative scope statements throughout so each section states what a command or validation establishes

Validation

• BASE_URL=/particleflow uv run --project .. jupyter-book build --html --strict
• pre-commit hooks on all changed files
• parsed the documented training, resume, and local Ray commands through the current CLI and typed configuration
• Hugging Face dataset dry run: 80 files, approximately 2.9 GB for the three CLD configuration-1 datasets
• Hugging Face model dry run: 79 files, approximately 222 MB for the published CLD checkpoint bundle

The full datasets were not downloaded and production-scale training and physics validation were not run. The examples use maintained command surfaces and existing integration paths.