feat(event cache): handle linked chunk updates and reload in the sqlite backend

[bnjbvr]

So, it's a bit of a chatty PR:

• because of the tests,
• because the tests required reading out from the store, I've implemented the full linked chunk reloading; it's using the intermediary raw format at this point, and later we can use the LinkedChunkBuilder to recreate the full thing.

Apart from that, the changes here are somewhat boring, and just running SQL requests.

I've taken the liberty to add a (clear) event id as a column for the events table, so we can add fetch_by_id easily in the future; this avoids a trivial migration.

Part of #3280
Split from #4308.

[Hywan]

Great work! I see no blocker, I just have a couple of questions and one feedback for a possible improvement. Other than that, great!

[Hywan]

I've just learned the difference between BLOB and TEXT in SQLite, but I wonder if we just not used a TEXT here. According to this forum thread:

BLOBs cannot be searched, compared, or manipulated.

I think we want TEXT here. Thoughts?

[bnjbvr]

Uhuh, interesting. That suggests we may have issues when there are multiple rooms, then 👀 I'll do a few experiments and change it, if needs be — pretty sure we've been using that in other places too.

[bnjbvr]

Pfew, https://sqlite.org/datatype3.html#sort_order says that BLOB are compared with memcmp, so we're fine. I've added a test that used the same chunk identifier in two different rooms, and it passes, so it's all fine as is. We're using BLOB in many many places, so if this should've failed, it would have failed elsewhere too ^^

So now, I do buy it that using TEXT would offer us more, e.g. searching substrings, prefixes, etc. I don't think it's a useful use case for us, to search for a room id substring, be it in the clear or hashed.

Thanks for raising this, though! TIL 😌

[Hywan]

Maybe we also want a TEXT here?

[bnjbvr]

Do you have a use case in mind where we'd need to search by prev_token?

[bnjbvr]

Related to the other comment: no need to specify TEXT unless we think we could need operations specific to a text object, which I don't think we need here either.

[codecov]

Codecov Report

Attention: Patch coverage is 97.17514% with 5 lines in your changes missing coverage. Please review.

Project coverage is 85.08%. Comparing base (3e7d7e8) to head (4b1e7e2).
Report is 31 commits behind head on main.

Files with missing lines	Patch %	Lines
crates/matrix-sdk-sqlite/src/event_cache_store.rs	97.12%	5 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4340      +/-   ##
==========================================
+ Coverage   85.01%   85.08%   +0.07%     
==========================================
  Files         275      275              
  Lines       30313    30487     +174     
==========================================
+ Hits        25770    25940     +170     
- Misses       4543     4547       +4     

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