Related T682

Related T682
Related P221

Related T329

Use case: indexers.  To send new tasks for scheduling to the
orchestrator queue (input queue), we need to check the mimetype and
license indexers queues (not only the mimetype one).

to streamline running experiments with different parameters

also, updated Rabin fingerprint parameters for better compression rates

down to 25m for linux.git, from 1h for the single process version, from 14h for
the sqlite3 version

instead of sqlite3

even if we're not using COPY, it is *much* faster

also, add driver shell script

Related T682

Related T682

This will help to filter the origins that should not be googlecode
ones but eclipselabs or apache-extras one

Related T864

Related T864

Related T864

For real huge data source (googlecode svn, indexers), using the
scheduler db is a no go (for now).

This leverages the same idea behind
swh.scheduler.backend_celery.runner without using the scheduler db.

This will permit to improve the rescheduling of T879 (googlecode svn).

Related T864
Related T879

Related T863

Related T863

Related T761

Related T761

- Update to be more resilient
- Permit to use loader-svn errors retrieval

Use case sample:

```
$ ./kibana_fetch_logs.py \
  --indexes "logstash-2017.09.18,logstash-2017.09.19,logstash-2017.09.2*,logstash-2017.09.30,logstash-2017.10.01,logstash-2017.10.02" \
  | tee svn-loader-error-logs-from-kibana.txt
```

Then:

```
$ cat svn-loader-error-logs-from-kibana.txt  \
  | ./group_by_exception.py \
  | jq ".googlecode"
{
  "errors": {
    "OSError(12, 'Cannot allocate memory')": 608
  },
  "total": 608
}
```

Related T676