n-gram character index

Currently the indexes are on token level:

• singleton index: [lemma] or [pos]
• pairwise index: [pos][lemma] or [pos][][pos]
• etc

Inspired by plocate, we could create character indexes instead (3-grams or 4-grams or ...). Where the characters are for lemmas, pos, etc, or combinations.

Instead of the entries [lemma="car"], [lemma="card"], [lemma="cart"], and [lemma="ward"] we would get:

• [lemma="*car*"], [lemma="*ard*"], [lemma="*art*"], and [lemma="*war*"]

This would only be interesting for some features (lemma, word, etc)

For combined features: [pos="ADJ"][lemma="*car*"], [pos="DET"][lemma="*art*"], [lemma="*war*"][pos="NN"], etc.

Start by reviewing the existing token-level index implementation and how queries for lemma, word, POS, and combined features are handled. Define which features and n-gram sizes are supported, then verify that the example character-index queries produce correct search results.