0.2.2

A release-pipeline fix only: the application, the database schema and the
configuration are exactly what 0.2.0 shipped, so there is nothing to migrate,
no new setting, and going back to 0.2.0 is safe. Nobody running the image needs
to hurry; the people it matters to are those who deploy to a Swarm with the
release workflow.

There is no 0.2.1 release. The tag was cut before its changelog entry and the
version bump had landed, so the workflow built and pushed an image and then
refused to publish empty release notes. A published tag is never moved; this
is the next number, carrying the same fix.

Changed

• A release that is not ready now fails before anything is published. The
  workflow checks, first of all, that the four package manifests carry the
  tag's version and that CHANGELOG.md has its section. Both used to be
  noticed at the last step, after the image was already in the registry — which
  is how 0.2.1 left an image behind with no release to go with it. A premature
  tag now fails in seconds with nothing built and nothing pushed.

Fixed

• The first automatic deployment no longer fails its own health check. The
  release workflow probed /health once, two seconds after creating the
  services, when a freshly created service has no update status for the
  convergence wait to look at — so on a stack's first deploy it asked before
  PostgreSQL and the application had started, and failed on a port nobody was
  listening on yet. It now polls within the same seven-minute budget and, if
  the budget runs out, prints the service's task list. Upgrades were never
  affected.

Container image

ghcr.io/liliang-dev/silencewatch:0.2.2