Feature: persistent headers

[psrok1]

Reasons to implement that:

• They make a good place for "properties" idea from RFC: Controllable task properties #167.
• It's also follow up from Upload switch to allow sharing samples with 3rd party services mwdb-core#801 as we don't propagate share_3rd_party header within karton-archive-extractor.
• Another painful header is quality that needs to be passed manually from task to task: https://github.com/CERT-Polska/karton-classifier/blob/master/karton/classifier/classifier.py#L102

Few design details:

• They follow the same semantics as persistent payloads, but both persistent and non-persistent headers are available from the same dict to keep it easier for users. Both internal and service code tends to directly modify self.headers collection so it was important also for compatibility.
• Copy of headers_persistent goes to the serialized JSON as well. Values contained there always take precedence over headers, so it should be consistent.
• For compatibility: we need to keep a copy in payload_persistent, because older consumer making a new Task(...) will pass headers without persistent ones so they will be lost. As we keep it in payload_persistent, they will be placed back in headers_persistent during deserialization made by karton.system