feat: improve retry backoff jitter

[polvalente]

I noticed that the current backoff uses an uniformily distributed jitter. However, if the random distribution is trully uniform, this means that half of the time the backoff finishes at half the delay it would be expected to finish (e.g. at an attempt with delay 5000ms, it has 50% probability to retry after 2500ms or less). This also introduces the problem that in some other percentage of the time the backoff is actually of linear or constant delay.

This PR refactors the function so the jitter is an additive noise proportional to the current delay, which should keep the backoff at the expected order of magnitude while adding the desired randomness.

A module option is provided to control said random noise level