Drop local workarounds when Tantivy 0.19.0 is released

• Drop deserialize_facet as the upstream impl of Deserialize has the same functionality then
• Replace CompoundNounTokenFilter by upstream SplitCompoundWords once that is released
• Collect query_terms directly into Vec<String> instead of going through BTreeMap<Term, bool>
• Use StopWordFilter::new(Language::German) for our de_stem tokenizer