Merge `training` and `examples`

[zafstojano]

Currently, we have two directories:

• training: contains verl / trl code used for training the models for our paper
• examples: contains minimal examples with various frameworks

@olliestanley suggested to label the state of the repo used for the conference submission, and move away from this distinction.

I would suggest we only keep the minimal training examples in examples, and update them as some are quite stale (e.g. 3-4 months old).

Ideally, they would all use the same training configuration (at least in terms of dataset generation), just show different implementation depending on the framework.