Fix migration timeout corruption

[jhunt]

It would seem that if we fail to run our knex migrations in the grace period that CF provides, we can corrupt the database, requiring manual intervention to fix.

The following does work, instead:

• Push the app without running migrations during .profile startup
• Run the migrations in a task

In order to keep the pipeline from hanging, we probably ought to change the health check from an "HTTP OK" to a "TCP port responds" check, and let Ghost 503 until we get migrations run.