inconsistent cellh5 file

[rhoef]

If a classifier was trained only on a subset of the feature categories (e.g. no spot features) and the spot features are turned on later on. The resulting cellh5 file is inconsitent.
That means exactly the definition part of cellh5 does not contain the spot feature while the feature table contains the data.

This means that the mapping of feature names and indices in the feature table is not correct any more.

[rhoef]

The good news is that the sanity check in the other direction is working i.e. if a feature is turned off later on an appropriate error pops up requesting a certain feature.