Investigation of missing notifications, particularly on iOS because it does not have a fallback

[biblicabeebli]

• see https://github.com/onnela-lab/beiwe-ios/issues/69 for current status.
• see https://github.com/onnela-lab/beiwe-backend/issues/402 for backend status
• see https://github.com/onnela-lab/beiwe-ios/issues/59 for ios discussion.

After an exhaustive review of the indicated study on our production server, which has a particularly high quantity of complex relative schedule surveys, I have determined that there is no bug on the backend. No anomalous push notification history was found.

Overview:

• There is logic that generates ScheduledEvents from Relative, Absolute, and Weekly Schedules.
• Absolute is trivial and correct.
• Weekly is called at a point that turns out to be brittle and should be updated.
• Relative is quite complex (they interacts with dates applied to a participant, they are called "intervention dates").

Some specifics I reviewed:

• All Schedule types must respect participant timezones and fall back to the default of the study timezone.
• There are several situations where all ScheduledEvents (either for the whole study or for a single participant) must be deleted and then generated anew.
• Newly generated schedules must check the history of the participant involved so as to avoid creating new ScheduledEvents for already-sent notifications (ArchivedEvents).
• The ArchivedEvent record should indicate schedules triggered at the correct time.
• Participant FCM tokens (required to send notifications, sourced from the device) should have a history to them that explains gaps and late notifications and notifcation groupings.
• When a device goes AWOL attempts at push notifications should continue for one week, at which point the FCM is marked as retired. (If a device appears again the normal process up checking in and providing the most-up-to-date-token resets the event.
• Manual push notifications should create correct records.
• Retired and deleted participants should not be sent notifications.

Other findings:

• There are now quite a few additional tests for event scheduling and retrieval.
• There was one minor bug that would result in a small number of excess notifications.
• There has been substantial code cleanup.
• There is an unknown scenario where weekly push notifications can fail to schedule a next notification, see https://github.com/onnela-lab/beiwe-backend/issues/401

(I want to change this language so both don't have the work schedule in them, but that is ... hard.)

[biblicabeebli]

• https://github.com/onnela-lab/beiwe-backend/issues/350 indicates that we need to update push notifications to not create items in the past..... maybe with the exception of first registration?

[biblicabeebli]

I have a v1 of backend code "finished" (there are substantial tests) but have not started on the iOS half of the work. This work is expected to be easier (drop in some string storage, send out some strings), but testing is likely to be difficult because we don't know exactly what the failures to receive push notifications actually were.