fix(pds): prevent relay desync after failed write (#162)

`applyWrites` was assigning the new `Repo` to in-memory state before sequencing the firehose event. If anything threw between then and `sequenceCommit` succeeding (e.g. mid-flight failure on an image post), Cloudflare rolled back the SQLite writes but the in-memory `Repo` stayed advanced. The next successful write then emitted a firehose commit whose `since` rev the relay had never seen, and the relay marked the repo desynced — requiring a manual `requestCrawl` to recover.

`this.repo` is now only assigned after the sequence + broadcast succeed in all four write paths (`rpcCreateRecord`, `rpcDeleteRecord`, `rpcPutRecord`, `rpcApplyWrites`), and any failure in that window invalidates the in-memory cache so the next access reloads from storage.

Author: ascorbic

.changeset/firehose-relay-desync.md

@@ -0,0 +1,9 @@
+---
+"@getcirrus/pds": patch
+---
+
+Fix relay desync after a failed write (e.g. an image post that errors mid-flight).
+
+`applyWrites` was assigning the new `Repo` to in-memory state before sequencing the firehose event. If anything threw between then and `sequenceCommit` succeeding, Cloudflare rolled back the SQLite writes but the in-memory `Repo` stayed advanced. The next successful write then emitted a firehose commit whose `since` rev the relay had never seen, and the relay marked the repo desynced — requiring a manual `requestCrawl` to recover.
+
+`this.repo` is now only assigned after the sequence + broadcast succeed, and any failure in that window invalidates the in-memory cache so the next access reloads from storage.