The Supervisor API expects a set of task descriptions to start, not the
actual tasks already started - as we were sending it.

This is still broken, since the consumers can't subscribe to the
producer if it hasn't started yet - and the app crashes.

See #2147

Instead of using consumers and producers, upon app startup we launch a
single SurveyCanceller process per each survey.

We still have to change the "online" behaviour (when we start cancelling
a survey while the app is already started).

See #2147

Instead of the old consumers and producers.

See #2147

Instead of having a producer and consumers, have a single actor per
Survey.

Still have to fix/refactor the canceller tests.

See #2147

Which may not necessarily be about SurveyCanceller, but this will do for
the time being.

See #2147

See #2337

The exit_code and reason are already being set by the SurveyAction

See #2337

We've dropped the "respondents" part of the name long ago, and it's part
of the runtime actually.

See #2337

See #2337

See #2337

See #2337

Looks like the log entries are getting stored in different order in the
CI, which is probably not really an issue.

This commit changes how we check for the last disposition change in the
tests, so we avoid asserting on a "response" entry log instead.

See #2337

Doesn't really have to do with the current PR, but are failing anyway

If cancelling a respondent raises an error, we log the issue in Sentry
and continue cancelling the rest of them.

We keep the survey Cancelling so we wait for human intervention.

See #2337

The function effectively marks the survey as terminated - the
cancellation is the whole process of managing respondents + the survey.

See #2147
See #2337

Co-authored-by: Ana Pérez Ghiglia <[email protected]>

See #2147
See #2337

Since it's shared in at least two different test modules.

See #2147
See #2337