crypto: Improve memory overhead of export_room_keys by using Vec::retain

[andybalaam]

As I understand it, Vec::retain should do the same job as into_iter().filter().collect() but potentially more efficiently. This could help our problem with running out of memory during export ( #2914 ).

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (3fea9ae) 83.47% compared to head (7034104) 83.47%.
Report is 11 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3013      +/-   ##
==========================================
- Coverage   83.47%   83.47%   -0.01%     
==========================================
  Files         222      222              
  Lines       23213    23212       -1     
==========================================
- Hits        19378    19376       -2     
- Misses       3835     3836       +1     

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

[bnjbvr]

Good catch, thanks 👍

It might avoid double-buffering for the entries of Vec<InboundGroupSession> indeed; I'd expect one issue is that get_inbound_group_sessions will still allocate all the sessions while we're discarding some of them. If get_inbound_group_sessions returned an iterator, we could filter out without even allocating the vector, which would be optimal.

[andybalaam]

If get_inbound_group_sessions returned an iterator

Yes, although I think it would need to be a stream because it's async. I may move on to doing this soonish, if it is needed to fix #2914