unicode normalisation form

[missinglink]

linked to pelias/api#1375

the icu_normalizer docs 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?