Let normalization work on all tensors

Orthogonal normalization only gets applied to Linear and 2D-Convolutional layers right now. We should investigate where we want them to apply. Is there a reason not to apply it to every kind of layer?

No files or tests are named. Start by locating the orthogonal normalization handling for Linear and 2D-Convolutional layers, then investigate whether other tensor or layer types have a reason to be excluded. Done means documenting the intended scope and implementing normalization for the agreed layer types with coverage for the new behavior.

Written by the indexing model from the issue text.