[pipeline] implement state merge logic in case of parallel processing or state restoration from destination

[rudolfix]

implement proper state management in Pipeline

1. state as visible by the source is just a dictionary (read and write via indexer)
2. each source iterator that requires state should operate on it's own namespace (Pipeline.state["namespace"] for explicit names or Pipeline.state for implicit names based ie. on name of calling module)
3. writing to state may only happen after iterator has stopped. that must be detected and extraction must be aborted
4. state is preserved atomically together with the results of the extraction
5. if many iterators operating on the same state were returned, we have a merge strategy: maximum value of all conflicting writes is preserved (ie. https://docs.meltano.com/guide/integration#internal-state-merge-logic)

Implementation

1. Consider locking state file when merging (ie. https://py-filelock.readthedocs.io/en/latest/index.html)