Expression graph autodiff extension

[TgDSML]

Summary

This PR extends the math_expr framework with explicit Expression Graph support and PyTorch Automatic Differentiation.

What's included

• Expression graph representation and parsing
• Conversion of expression trees to torch.nn.Module
• PyTorch node implementations
• Executor adapter for integrating expression graphs with the existing framework
• Unit tests
• Example scripts demonstrating parsing, visualization, execution, and automatic differentiation
• Experimental scripts used to validate the implementation
• Documentation for the extension

Validation

• All source files compile successfully.
• All tests pass.
• Example and experiment scripts execute successfully and generate their outputs automatically.

Generated outputs (figures, metrics, etc.) are intentionally not tracked, as they can be reproduced by running the included scripts.