Add dense PCA fallback and align sample projection tests

[SauersML]

Summary

• add a force_dense_pca configuration flag and let compute_pca short-circuit into a dense SVD pipeline when it is enabled
• implement a compute_dense_pca helper that performs the dense PCA using f64 SVD outputs and returns the expected EigenSNP structures
• update the sample projection parity tests to request the dense path, derive Python references from the training slice, and align component signs before comparing scores

Testing

• cargo test --release

---

https://chatgpt.com/codex/tasks/task_e_68d40be0dd44832ebe44d50cfc90d1af