Edge n-gram autocomplete

Pre-compute prefixes at index time: "nike" is stored as "ni", "nik", "nike", so each keystroke is an exact token lookup. The search analyzer only lowercases, or "nik" would also match every "ni" word.