Ensure initialization signals get fired during sync

[rtibbles]

Summary

• The default transfer_stage value set for contexts is INIITIALIZING - because of this, the started signals for this stage are never fired, as the middleware processing always flags it as at that stage.
• To resolve this, I've added a manual firing of the initializing started signal when a transfer session object is created
• Adds a test to ensure this is the case.

TODO

• Have tests been written for the new code?

Reviewer guidance

I initially attempted a more general approach by creating a NO_STAGE stage to allow the initializing started signal to get triggered, but it broke another test. While it fixed the test here, it did not seem to alter the behaviour in Kolibri. I also didn't know whether this would have implications for the Kolibri sync commands, which seem to explicitly fire the initializing started event: https://github.com/learningequality/kolibri/blob/release-v0.16.x/kolibri/core/auth/management/utils.py#L615

Issues addressed

This should resolve learningequality/kolibri#10588 by ensuring that when an inbound sync happens, the pre-transfer hooks are properly executed: https://github.com/learningequality/kolibri/blob/release-v0.16.x/kolibri/core/auth/sync_event_hook_utils.py#L130

[bjester]

This shouldn't be required. The issue must exist in Kolibri still

[rtibbles]

Closing in favour of learningequality/kolibri#11268

I do find it a little counterintuitive that the initializing stage doesn't fire a started signal - but also agree that if it's going to happen in Morango, it should happen for both server and client.

[bjester]

@rtibbles It is indeed counterintuitive, I agree. I think I might have added the signals before the chain-of-responsibility architecture, so we didn't have the context classes whereas before everything revolved around the sync and transfer model objects (which don't exist at that time obviously).

Firing the started signal in a standard way would need to be handled in the controller. There's some awkwardness there with the existing sync clients though. We'd also need to ensure that Kolibri's signal->hook parsing is defensive against missing sync and transfer session objects