fix(syncer): stop backfill from regressing latest_message_id

[hannesrudolph]

Fixes #85.

Problem

syncBackfillPages persisted the newest id of each backfilled page to channel:<id>:latest_message_id. During a resumed --full sync the backfill region is by construction older than the stored head, so every page write regressed the pointer. When the per-channel timeout then deferred the channel, the next pass's forward sync re-crawled the entire span between the backfill cursor and the channel's true head (all duplicate INSERT OR IGNORE work) before backfill got whatever remained of the timeout budget — so resume cost grew roughly quadratically with channel size. On a ~200k-message channel, net-new rows per 5-minute pass collapsed from ~28k to ~500.

Fix

syncFullChannelHistory already knows the current head (state.Latest maxed with the forward-sync result), so it now passes that into syncBackfillPages as a floor. The in-loop channelLatestScope checkpoint is only written when the backfill page's newest id actually advances past the floor — which only happens in the one case the write existed for: a first-run --full whose backfill starts at the channel head (no stored latest yet). With a stored head in place, backfill now only advances channel:<id>:backfill_before_id, and the pointer can never regress.

Test

TestSyncFullBackfillDoesNotRegressLatestPointer reproduces the issue: a half-backfilled channel (latest=250, backfill cursor=200) whose backfill is interrupted by context.DeadlineExceeded mid-run. Without the fix the latest pointer regresses to 199 (the newest id in the backfilled page); with the fix it stays at 250, and the resumed pass fetches only the remaining 99 backfill messages instead of re-crawling the head span.

Verified: go build ./..., go vet ./..., go test -count=1 ./..., and go test -race ./internal/syncer/ all pass; the new test fails against main with expected: "250" / actual: "199".

🤖 Generated with Claude Code

[clawsweeper]

ClawSweeper status: review started.

I am starting a fresh review of this pull request: fix(syncer): stop backfill from regressing latest_message_id This is item 1/1 in the current shard. Shard 0/1.

This placeholder means the worker is alive and reading the current context. I will edit this same comment with the actual review when the claws are done clicking.

Crustacean status: shell secured, claws on keyboard, evidence pebbles being sorted.