Structured JSON output + other minor fixes

[lfoppiano]

In this PR I've added the JSON output that provide a structured output format:

{
  "question": {
    "title": "In which language should the questions be posted? Portuguese or English?",
    "body": "In which language should the questions be posted? Portuguese or English?\nIf a question is posted in a language different from the expected, what should we do? Flag to close or translate and edit?\n"
  },
  "answers": [
    {
      "id": "5",
      "body": "I think all contributions (questions, answers and comments) should be allowed in portuguese and\/or english.\nIn the case of a question or an answer, you can always edit it to add a translation. In the case of a comment you can not edit to translate it, but the comments that add value to a question or to an answer could be included in such question or answer, using both languages if appropriate.\n",
      "score": 13
    }
  ]
}

In fact this format is used a model and it's serialised in different ways depending on the output format.

There is also an additional flag --keep-sources that will not delete the 7z files in the dump directory

[lfoppiano]

This change requires the latest version of https://github.com/lfoppiano/lm_dataformat

[lfoppiano]

This requires updating lm_datasource as well, which currently is required in edit more