Add option to re-train vosk models on certain words

Vosk offers the option to update the recognizer dictionary at run-time: https://alphacephei.com/vosk/adaptation

This would be useful to correctly recognize special vocabulary users want based on their transcript corrections. We should look into if/how we can integrate this in a nice way.

Some Challenges I see
* We need to store information on the updated vocabulary.
* What do we need to store?
* Do we store it on the server? On the clients?
* If on the clients, do we store it in the audapolis files?
* If on the server, do we need accounts? And options to manage different vocabularies?
* How to we add new words to the vocabulary? Automatically? Does the user have to choose? Is it a seperate option? Do we ask them if they want to add a word if they want to add it after they entered it X times?
* We probably need to re-transcribe at least part of the source media to get the better transcription from the updated words. When do we do that? Background and automagically update parts of the document while the user is not looking? Syncronously and pause the users workflow for that? Do we let the users when choose to retranscribe? Do we offer this per paragraph? Per document? Per source media? If per paragraph, how to we handle paragraphs that do not consist of one continous piece of source media?