I have a use case for Celery [abstract
tasks](https://docs.celeryproject.org/en/latest/userguide/application.html#abstract-tasks)
to handle exceptions for an entire Celery app in a common way. Handled
exceptions should be swallowed in the abstract task and never make it to
Sentry.

Prior to this change, the Celery integration always instruments
`task.run` and incorrectly instruments `task.__call__` (`task(...)` is
equivalent to `type(task).__call__(...)`, not `task.__call__(...)`).
After this change, we'll use the same logic as Celery to decide whether
to instrument `task.__call__` or `task.run`. That change allows abstract
tasks to catch/raise exceptions before the Sentry wrapper.