[callumalpass]

Fixes PERF-19 and most of PERF-5 from the overnight perf report.

Problem. The SDK records a durable receipt after every applied action. If the state store implements appendJournal, it appends one small event; otherwise it rewrites the whole mirror state, including every prepared document in the batch. The plugin's IndexedDB store never implemented it, so a first sync's writes grew quadratically with the number of notes.

Fix.

• IndexedDbMirrorStateStore appends events under [key, "journal", n] in the existing object store and replays them on read with the SDK's own applySyncJournalEvent. Every full write and clear drops them in the same transaction.
• The checkpoint summary is unaffected.
• ReceiptObservingStateStore forwards appendJournal (the SDK detects it by presence) and reports history from journal events, including for a batch resumed after a restart.
• The SDK patch exports applySyncJournalEvent and SyncJournalEvent from the mirror entry; a matching upstream PR follows.

First sync (fake-indexeddb)	Before	After
250 notes	508 puts, 190 MB, 664 ms	256 puts, 1.4 MB, 80 ms
1,000 notes	2,008 puts, 3.0 GB, 11.9 s	1,006 puts, 5.7 MB, 385 ms

On real IndexedDB the "before" side is worse: every write is structured-cloned and flushed to disk.

Tests (test/indexedDbJournal.test.ts, real SDK engine and real store):

• full-state writes stay constant as the vault grows, and bytes grow linearly;
• a crash mid-batch resumes from state plus journal in a fresh adapter;
• full writes and clears drop the journal atomically;
• out-of-order and baseless journals are reported as corrupt;
• history reports every file exactly once across a restart.

Full suite: 434 pass, plus 10 lossless round trips; lint and mobile check pass. No backwards compatibility is needed: unreleased.