Refactor to feed data in chunks

[maurolepore]

We already know large lonabooks crash people's computers and already found a way to reduce the chances:

https://rmi-pacta.github.io/r2dii.match/articles/chunk-your-data.html

I would also consider a line-by-line process which is not described in that article. That's the approach I take in tiltEp sometimes. It takes longer to process but has no upper limit for how big the loanbook can be.

[maurolepore]

For an enhanced approach see https://2degreesinvesting.github.io/tiltIndicator/articles/handling-long-runtime.html