[FEATURE] Compute the input/output tokens of a dataset

[plaguss]

Is your feature request related to a problem? Please describe.
After merging #1034 we will have available the statistics generated by an LLM. We could add a functionality to compute some statistics on these variables.

Add an easier way of passing functions that can operate on these generated statistics to make simple doing computations on the statistics generated.

Describe the solution you'd like
A report in the README of a dataset with the number of tokens generated, and some descriptive stats per row for example.

Describe alternatives you've considered
Compute this by hand.

Additional context
Add any other context or screenshots about the feature request here.