Investigate memory usage and an on-disk index

Even on small projects the language server frequently uses several gigabytes of memory, especially when there are lots of symbols to be indexed (note that this includes the entire dependency tree!).

Since we probably do not query the database so frequently that we _need_ it to be in memory, perhaps we could look into using an on-disk database such as SQLite? (This would additionally have the advantage of not requiring a reindex every time the project is opened, which would be in line with #337)