Related to T2367

- blackify all the python files,
- enable black in pre-commit,
- add a black tox environment.

- Rename --max-messages to --stop-after-objects
- Drop the explicit loop within the current implementation and let the base one
  do the job

Following the journal migration (v0.0.27).

msgpack now defaults to raw=False.

This commit preserves compatibility with older versions of msgpack.

This fixes a sphinx warning as it is also used in swh-icinga-plugins.

Related to T2188

Else, 'http://foo.foo/foo' has a better score than 'http://foo.bar'
when searching for 'foo bar'.

It makes origin update very slow, and was only useful for tests.

Instead, this commit adds a 'flush' endpoints that is called by
tests, because they need the indices to be refreshed to test
the results of a search query.

Sorting on the document id causes Elasticsearch to error
because it uses too much memory.

Allows us to call tests on things other than python3, with the same settings.

For uniformity with the indexer-storage.

This simplifies the Docker entrypoint (it can just call 'initialize'
unconditionally); and allows easy addition of new fields to an index.

Field removal would still need to be done manually.

Origins are no longer dicts.

The journal client now uses Consumer.consume instead of Consumer.poll,
so not mocking Consumer.consume resulted in infinite loops.

thanks @vlorentz for pairing with me for this!

This also adds a 'has_visits' boolean field to origin documents,
which will be set to True as soon as a visit for the origin is
seen.

'cursor' has a different meaning in our components using postgresql;
this renaming will avoid confusion when adding token-based pagination
to these components.

It keeps failing and I don't know what it is.

Related D2082