Queue and retry to-device msgs, so E2EE works in bad networks or with unreliable HS

[kittykat]

Queue the to-device messages. We should make sure that we retry to-device messages. New implementation on web saves them in a queue to persistent storage to be sent when they can be.

These issues must be resolved for this issue to be considered done:

Tasks

Preview Give feedback

1. We don't retry E2EE to-device messages at all. vector-im/element-web#12851
   A-E2EE O-Uncommon S-Major T-Defect Z-Chronic +5
   
2. Retry to-device messages matrix-org/matrix-js-sdk#2549
   T-Enhancement +1
3. Retry to-device messages vector-im/element-android#6689
   A-E2EE O-Frequent S-Major T-Enhancement Team: Crypto +5
4. Retry to-device messages matrix-org/matrix-ios-sdk#1538
   +0

[dkasak]

Should be noted that matrix-org/matrix-js-sdk#2549 only retries room key (Megolm session) sending; other to-device messages are still not retried. This is tracked in matrix-org/matrix-js-sdk#1866.

[jdauphant]

@giomfo maybe we can push this ticket to solve part of our keys sharing problems.

[pmaier1]

According to @BillCarsonFr, this will be solved for other platforms once we have Element R integrated which will happen soon.
The web-specific part (#673 (comment)) will also be fixed by integrating the rust SDK.

[pmaier1]

@BillCarsonFr Can you close this and the related tasks if they are solved by Element R? Thanks!