Lipschitz top-n truncation for persistence diagrams

Construct a genuinely 1-Lipschitz replacement for top-$N$ persistence-bar truncation under the bottleneck metric, while retaining a computationally manageable representation for the COMPLEX embedding.

Background

The implementation retains only the 50 most persistent bars in each slice. The paper notes that this truncation is not 1-Lipschitz for the bottleneck metric, so the stated guarantees apply to the truncated fibered barcode rather than automatically to the untruncated module. A soft truncation that is genuinely 1-Lipschitz would remove this gap and provide a cleaner stability relationship between the computed descriptor and the full persistence information.