Improve SVD pipeline

[DSilva27]

• [ ] Use randomized SVD to reduce computational time (plus, the volume matrix is low rank)
• [ ] Instead of projecting all volumes to the same subspace, project each submission into it's own subspace. Compare submissions by computing the angle between subspaces.
• [ ] Keep the possibility of choosing a reference submission (or GT) and project all submissions there.
• [ ] Allow the user to specify masking the volumes
• [ ] Allow the user to choose a b-factor for b-factor normalization