Fix novel on cells without counts in the hvg peaks

[rcannood]

In run_2026-09-28 novel_predict failed on the multiome swap datasets with a missing tensor.pt. On bmmc_multiome/swap, 3092 cells have no counts in the 1000 hvg peaks the LSI uses, so the TF-IDF divided by zero and the LSI was NaN for those cells. The validation RMSE was then never better than the initial 100000, so train_and_valid() never saved a model, while novel_train itself still exited 0.

Describe your changes

• novel: treat an empty row as having a row sum of 1 in the TF-IDF, and a zero standard deviation as 1 when scaling the LSI.

On the de.NBI copy of bmmc_multiome/swap, the LSI goes from 791552 NaN values to none.

Checklist before requesting a review

• I have performed a self-review of my code

• Check the correct box. Does this PR contain:

  • Breaking changes
  • New functionality
  • Major changes
  • Minor changes
  • Bug fixes

• Proposed changes are described in the CHANGELOG.md

• CI Tests succeed and look good!