feat(sdk): Rotate session keys when a member leaves the room

[kaylendog]

This PR adds an event handler on start that discards (and hence rotates) the current session whenever we see a leave membership event.

Part of element-hq/element-meta#3078

• I've documented the public API Changes in the appropriate CHANGELOG.md files.
• This PR was made with the help of AI.

Signed-off-by: Skye Elliot [email protected]

[codspeed-hq]

Merging this PR will not alter performance

✅ 50 untouched benchmarks

---

_{Comparing kaylendog/history-sharing/respect-visibility (177f38a) with main (a4bb2f3)}

[kaylendog]

I'm tempted to use unwrap here, but I am wary that if any of these assumptions are wrong, this will crash the client entirely.

[codecov]

Codecov Report

❌ Patch coverage is 65.21739% with 8 lines in your changes missing coverage. Please review.
✅ Project coverage is 89.90%. Comparing base (d6c666a) to head (177f38a).
⚠️ Report is 30 commits behind head on main.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
crates/matrix-sdk/src/encryption/mod.rs	63.63%	6 Missing and 2 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #6292      +/-   ##
==========================================
- Coverage   89.94%   89.90%   -0.05%     
==========================================
  Files         374      374              
  Lines      102613   102635      +22     
  Branches   102613   102635      +22     
==========================================
- Hits        92299    92270      -29     
- Misses       6771     6799      +28     
- Partials     3543     3566      +23     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[richvdh]

@kaylendog in the description of the PR, could you link the issue that explains why we need to do this?

[richvdh]

mostly lgtm, a couple of small comments

[richvdh]

do things added by add_event_handler definitely get called when state changes outside the timeline section of the sync response?

It might be good to do a test which does something like:

• Charlie leaves the room
• Alice sends Message 1
• Alice goes offline for a while
• Bob invites Charlie
• Charlie joins
• Charlie leaves
• Bob sends a bunch more messages
• Alice comes back online
• Alice sends Message 2

The point here is that, from Alice's point of view, Charlie is in the leave state at both Message 1 and Message 2, and she'll never even see the invite/join/leave stage unless she backfills. She should still discard the session because she'll see a different leave event for Bob, but it's a bit subtle.

[kaylendog]

This exact scenario did occur to me, but I want to avoid walking back through all the state changes up until we sent the last message... I'm hoping the handler fires for us here - I'll test.

[kaylendog]

I've implemented a test for this in 4a93def which should be passing, but I do think it's worth discussing the potential sliding sync a bit further.