Increase atomicity of the OCR task

Right now the OCR task works on an entire document version at a time. This feature would split the OCR task to be a task for each document page, allowing to better spreading the load among workers. Would also allowing for easier retries of failed page OCR without having to resort to state keeping.