Fixes the #82/#136 lock-up: the background scheduler held a Postgres connection while parked, and the plugin is re-instantiated on every Dispatcharr discovery (UI plugins-list polling), so connections leaked until max_connections was hit and the server locked up. Independent of channel count, so it affected small installs too.
The scheduler now closes its DB connection before every sleep and on exit, and Plugin.init is idempotent (leaves a healthy scheduler running instead of churning a new thread per discovery). Verified live: connection count stays bounded and login no longer wedges. See CHANGELOG.md.