Add FACTM to the sample representation methods

[VladimirShitov]

Description of feature

Małgorzata Łazęcka has recently presented a FACTM model: https://github.com/szczurek-lab/FACTM/tree/main . While conceptually similar to MOFA, it contains a rotation method for latent factors. Combining it with filtering out batch effect-related factors (see #102) could provide good sample representations. Would be great to see this method in patpy

How to contribute

1. Fork the repository
2. Implement a new class for FACTM in https://github.com/lueckenlab/patpy/blob/main/src/patpy/tl/sample_representation.py

• It must be inheritor of the SampleRepresentationMethod class
• Take a look at pseudobulk for a basic example or at MOFA for a more comprehensive example
• The class must contain 3 methods:
  • __init__() to set it up
  • prepare_anndata(), with setting up the data and the model
  • calculate_distance_matrix(), which outputs a distance matrix between samples

3. Open a pull request to merge the code to main