The SQLite tokeniser does not deal with scripts that do not use spaces
for word breaking (CJK, Thai, etc), thus searching in those languages
does not work well.
This adds a custom SQLite tokeniser based on ICU that breaks words for
all languages supported by that library, and uses NFKC_Casefold
normalisation to handle normalisation, case folding, and dropping of
ignorable characters.
Fixes#121
Add SQL migration that drops the old FTS4 MessageSearchTable table,
re-create as a FTS5 table, clean up the column names a bit, and adds a
flags column so unread/starred queries can be made fast.
Define a SQLite FTS5 extension function `geary_matches()` to replace
the FTS3 `offsets()` function which no longer exists in FTS5, based on
Tracker's implementation.
Update code to FTS5 conventions (docid -> rowid, etc), use new column
names, populate and update the flags column as the email's flags
change, and use new match function for getting matching tokens.
Advanced searches are probably currently broken, these will be fixed
by subsequent commits.
