[FEATURE] Bias modelling

[Livia-Rasp]

Is your feature request related to a problem? Please describe.
At the moment the LFM has to account for all the variation in the data.
It could however improve the data imputation to add a bias, which accounts for sample and feature specific effects so that the LFM would only need to account for the effect of the interactions of samples and features.

Describe the solution you'd like
As described by Koren et al
Bias could either just be row and column means or also be trained during the GD

Additional context
It could also be interesting to return the bias to recieve some feedback. Maybe also for further analyses.