Implement new Standford's paper TextGrad in ax

[gunta]

The new Stanford's paper is very interesting, all optimizations on the paper get a better accuracy benchmark over DSPy.

The paper
https://arxiv.org/pdf/2406.07496

The code
https://github.com/zou-group/textgrad

A nice introduction
https://www.youtube.com/watch?v=Qks4UEsRwl0

[dosco]

thanks have seen this paper couldn't figure out what it does besides the focus on making a pytorch like api. Where does the accuracy
improvement come from?