This also tries to type a tad better the indexers.

This will allow to send metrics at some point.

This allowed to chain indexers together.
It's not used, so to simplify maintenance, remove it.

This also clarifies the intent.

Similarly to what we do in the swh.storage.

Related to D2580

`storage.origin_get` can return None as origin. This adds the behavior to skip
those None origin instead of failing the indexing task.

Related https://sentry.softwareheritage.org/share/issue/8551b4a0ed694e09ad1ad7bde047e48f/

This should improve the textual content reading queries.

Related to T1788

In production, the previous query version takes too long to finish. This ended
up with timeouts (with query still running in the pg backend). Accumulating
similar new queries.

Counter-intuitively, specifying twice the range query to use (both in
content_mimetype and content_fossology_license), the execution now returns
immediately.

Close T1788

And re-raise them through the RPC layer.

This is now required by swh-core, as exceptions are no longer pickled.

Required by swh-storage >= v0.0.172.

Also removes types from docstrings, where unnecessary.

Closes T2258

get_storage functions are usually imported `from X import get_storage`,
so monkey-patching `X.get_storage` has no effect.
This commit reverts the behavior change introduced in
c018eeaa.

Also fixes bugs like: `TypeError: indexer_configuration_add() missing 2
required positional arguments: 'self' and 'tools'` since swh.core
0.0.87.

Instead of directly import the storage backend whose name could change.

This fixes the ci build [1]

[1] https://jenkins.softwareheritage.org/job/DCIDX/job/tests/817/console

Related to T2188

Removing these methods causes the indexer to fall back to
the default (_try_get_head_generic), which already does
the right thing.

It must be a Person (or Organization), not Text.

Adding a '.' in a description field did not please a couple of tests, so revert
to the original desc, but use " instead of ' to please codespell.

This allows you to run tox -e py3-slow, tox -e py3-prop, tox -e py3-prop-slow,
tox -e py38-slow, tox -e pypy, ...

They were running on the postgresql storage instead.

Origins are no longer dicts.

- extract a bunch of initialization/testdata in a conftest.py
- add swh_indexer_storage and swh_indexer_storage_with_data fixtures
- 'explode' the main CommonTestStorage class into several pieces (one for
  every 'entity type' dedicated endpoints, eg. content_mimetype, ctags, etc.
  plus one for misc endpoints),
- get rid of the gen_generic_endpoint_tests() helper function, just write
  these test functions and inject them in endpoint_type specific test classes,
- use these classes + the fixture based client/server test framework from
  swh.core for test_api_client,
- use these classes for test_in_memory.

replace the 'start' argument by a 'page_token' which is now expected to be
an (not so) opaque token used for pagination.

The 'end' argument is dropped.

The method also returns a dict now (instead of generating the origins) with
2 keys: `next_page_token` and `origins`.

Adapt tests and the list_origins_by_producer() helper function (in cli.py)
accordingly.