Improve scripts for profile computation: searching from (serialized) indexed trees, and using ranges to compact occurrences found

Add types to represent indexed sentences and documents, effectively absorbing part of the work that was done during the search

Add a script to serialize syntax trees computed from the .conllu files, copying part of the Conllu.Type types in the process to get a Generic instance for Serialization and to slightly improve structure by indexing features by name with a Map

Lift some constraint on the monad in  and use that to pass the config around as a Reader in the new profile extractor script