speedy indexing / avoiding troubles
it shouldn't be needed to do a 100% full index build, doing and continuing a partial index run must be possible (STILL MISSING)
one idea to implement partial indexing / resumable indexing could be to have different indexing phases:
index all items' metadata, but not the content - this should be rather quick.
index the current revision of all items, but not the content.
index content of current revisions
index content of all other revisions
Problems that can happen: we have to expect crashes from malfunctioning filters we have to expect misbehaved converters/filters (like consuming and not freeing lots of memory)