send_message silently fell back to the default conversation whenever the
plan-time or slack-channel lookup errored, because the query error was
ignored. A transient "database is locked" (from the channel-creation/sync
routine holding the write lock while making slow Slack API calls) thus
misrouted a service notification to the admin DM instead of the event
channel.
- api.go: capture lookup errors and only treat gorm.ErrRecordNotFound as
"no service / no channel"; on any other error, fail so the caller retries
rather than posting to the wrong conversation.
- database.go: open SQLite with WAL journaling and a 10s busy timeout so
reads proceed alongside the sync writer instead of locking.
- Bump version to 0.2.2.
