Key import race causing a worse key to be saved in place of a better version.

As seen in this rageshake https://rageshakes.element.io/api/listing/2024-11-06/130841-YOSPUGGB/console.2024-11-06-12.log.gz

There is slightly late room key, so there is an UTD and the sdk tries to query the key from key storage.
And just after the room key arrives down the sync.

We then have two concurent actions trying to import the key

• From key storage (using import_room_keys)
• And down the sync (using handle_key)

The race is that they both try to see if there is an existing key in order to see if the new one is better or worse

https://github.com/matrix-org/matrix-rust-sdk/blob/8d07f36247e92c5cfae4c8d0b60f6ea56743bcdf/crates/matrix-sdk-crypto/src/store/mod.rs#L1748-L1756

and

https://github.com/matrix-org/matrix-rust-sdk/blob/8d07f36247e92c5cfae4c8d0b60f6ea56743bcdf/crates/matrix-sdk-crypto/src/machine/mod.rs#L891-L906

But they both think they are the first and that there is no matching key in store.

Therefore the first one to commit to the store wins. By-pasing the better/worse check.
The late key is better than the key from key storage, because it is safe (no gray shield)

Start with import_room_keys in crates/matrix-sdk-crypto/src/store/mod.rs and handle_key in crates/matrix-sdk-crypto/src/machine/mod.rs, focusing on their existing-key checks and store commits. Reproduce or trace concurrent imports from key storage and sync, then verify that the better late room key is retained rather than whichever import commits first.