[Demo] Add GAOT (Geometry-Aware Operator Transformer) example

[Lithiumcr]

Core features:

• Implement complete GAOT architecture (MAGNO Encoder + Patch ViT + MAGNO Decoder)
• Add AGNO (Attentional Graph Neural Operator) with attention mechanism
• Add GeometricEmbedding (statistical and pointnet methods)
• Support multi-scale feature processing (scales=[1.0, 0.5, 0.25])
• Add neighbor search (radius/knn with caching)
• Add scatter operations (replace torch_scatter)

Framework integration:

• Seamless integration with PaddleScience framework
• Support both JSON and YAML configurations
• Complete train/eval/export/inference pipeline
• Standard ppsci.solver interface

Code quality:

• Apache 2.0 license headers added to all files
• 13 Python files, ~4,301 lines of code

Known issues (to be fixed in follow-up PRs):

• Projection layer dimension issue in MAGNODecoder
• End-to-end forward pass test pending
• Case documentation (docs/zh/examples/gaot.md) to be added

Reference:

• Original PyTorch implementation: https://github.com/Shizheng-Wen/GAOT
• Original license: MIT
• Paper: Geometry Aware Operator Transformer (NeurIPS 2025)

Co-authored-by: Original work by Shizheng Wen and team

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

lichangrong seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[paddle-bot]

Thanks for your contribution!