Incremental operation support

[vmarkovtsev]

EPIC: src-d/backlog#1435

identity-matching already caches the raw input data. In order to be able to run continuously, we still miss the following:

1. Running loop.
2. Generating diff to the previous results.
3. Running src-d/parquetry as a service.
4. Supporting diffs in src-d/parquetry: this means supporting removing records.