unicode normalisation form

linked to https://github.com/pelias/api/issues/1375

the [icu_normalizer docs](https://www.elastic.co/guide/en/elasticsearch/plugins/current/analysis-icu-normalization-charfilter.html) say that the default normalization mode is `nfkc_cf`.

this is basically `nfkc` with 'code folding' too, so we could switch to this mode so we don't have to have a `icu_folding` filter step?