feat: add HF-backed checkpoint storage

[SoheylM]

Summary

• add a shared checkpoint backend with HF, W&B, local, and auto resolution paths
• route training entrypoints through the new checkpoint packaging flow while keeping W&B compatibility
• update evaluators and surrogate-model loading to prefer HF packages with W&B fallback
• document the new workflow and add a migration playbook for historical W&B checkpoints

Validation

• conda run -n engibench python -m py_compile on modified Python files
• conda run -n engibench ruff check on modified files
• conda run -n engibench mypy --follow-imports=skip engiopt/checkpoint_store.py engiopt/surrogate_model/run_pe_optimization.py engiopt/vqgan/evaluate_vqgan.py
• local checkpoint package resolution smoke test

Follow-up

• run authenticated smoke tests for HF-backed save and legacy W&B restore before review

[SoheylM]

@mkeeler43 requesting your review on this one. This PR adds an HF-backed checkpoint backend for EngiOpt so new model weights can live on Hugging Face instead of saturating W&B artifact storage, while keeping backward-compatible restore support for existing W&B-hosted checkpoints. It also packages run config and metadata with the checkpoint so new HF-backed runs remain reproducible and tied back to the originating W&B run.