Taking a look to the current implementation for the Spellchecker, the first thing I noticed is that is a little bit spread across Solr and Lucene.
Actually it is not easy to place the issue in one or the other.
As a first observation it should follow the AutoSuggest structure, so it should be in Lucene.
In particular this issue reflects the requirement of having the possibility of a Spellchecker queryTime/ index Time aware of the analysis.
Currently the analysis is applied in Solr at query time , but it is not possible to configure the analysis that is going to be used to then build the Index time FST used for matching the spell corrections.
We should discuss a design ( I know it would be quite a big work to refactor the spellcheck module, so maybe we are not going to do big changes but simply implement the requirement)