Fix indexeddb reopen connections

[zzorba]

Summary
Every IndexedDB store (event_cache, media, state, crypto) holds a single long-lived IDBDatabase handle, treats close()/reopen() as no-ops, and has no connection-close listener. When the browser closes the connection out from under us — a transient storage-backend error, eviction, or a versionchange from another browsing context — every subsequent transaction fails for the lifetime of the process, so the store can neither read nor write until the page/app is restarted.

On a long-lived desktop (Electron) client this surfaced as timelines universally failing to load events and synced messages being permanently dropped (recoverable only by logout/login). For the event cache it's especially damaging: persistence runs on a decoupled task whose errors are only logged while the sync position keeps advancing, so events dropped during the dead window are never re-requested.

This makes the connection handle recoverable across all four stores.

Shared mechanism (src/connection.rs)
IndexeddbConnection wraps the handle as Rc<RefCell<Rc>> and exposes a generic with_transaction(stores, mode, open, body) that builds the transaction and, on a DOM InvalidStateError (connection closing/closed), reopens via open and retries exactly once. The owning Rc is kept alive on the helper's stack frame for the duration of body, so the borrowed transaction is sound — no unsafe, no background task (async closures / AsyncFnOnce).

Per store

• event cache — all transaction sites + reopen().
• media — all transaction sites + reopen().
• state — all transaction sites (sync token, room infos, room/stripped state, profiles, receipts, account data, send queue, dependent send queue, thread subscriptions, custom values) + reopen(), reopening via upgrade_inner_db using the retained store cipher and meta db. The meta db (migration bookkeeping/backups only) stays a plain handle.
• crypto — all transaction sites (account, identity, sessions, inbound/outbound group sessions, devices, gossip/secret requests, backup keys, withheld sessions, room settings, key bundles, custom values, lease locks) + reopen(), reopening via open_and_upgrade_db with the retained serializer.

Also adds From<indexed_db_futures::error::Error> to the event-cache and media error types to satisfy the shared helper's error bound.

This is the connection-lifecycle layer. Natural follow-ups (not included): typed transient-vs-logical error classification, and a recovery path that re-fetches a batch dropped during the dead window rather than relying on the next sync.

Test plan
cargo check -p matrix-sdk-indexeddb --lib --target wasm32-unknown-unknown passes for all four stores (they are #[cfg(target_family = "wasm")]).
Mechanical audit: zero remaining direct self.inner.transaction(...) sites; every operation routes through the reopen-and-retry helper.

• I've documented the public API changes in the appropriate changelog files (see Writing changelog entries).
• This PR was made with the help of AI.

Signed-off-by: Daniel Salinas

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 89.88%. Comparing base (d2f7ce4) to head (51bb961).
⚠️ Report is 23 commits behind head on main.
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #6675      +/-   ##
==========================================
- Coverage   89.89%   89.88%   -0.01%     
==========================================
  Files         396      396              
  Lines      110236   110236              
  Branches   110236   110236              
==========================================
- Hits        99098    99089       -9     
- Misses       7376     7383       +7     
- Partials     3762     3764       +2     

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

[codspeed-hq]

Merging this PR will not alter performance

✅ 50 untouched benchmarks

---

_{Comparing filament-dm:zzorba/fix-indexeddb-reopen-connections (51bb961) with main (e8673e4)}

[Hywan]

Thank you for your contributions.

+2'312-1'777 for 5 commits. Is it possible to split the commits into smaller ones, please, to facilitate the review process?