Description

Using the current implementation of "migrations" causes a heavy load on the database as it locks the tables during "create if not exists" while starting a worker process. This would be fine if only a single worker is used and there is no concurrent access to the database but in most scenarios you would use a WSGI server like gunicorn which spawns multiple workers which all try to apply migrations at the same time and lock the tables. Gunicorn may also restart a worker if it has not accepted a request in a specified timeout which will resurrect and try to apply migrations, thus locking the tables and all current requests going to other workers. The requested implementation will not lock tables every time and it will only do so if there are unapplied migrations. This also helps with updating the database schema obviously and there would not be need to check if a field exists like:

BEGIN
    ALTER TYPE history ADD ATTRIBUTE "user" text CASCADE;
EXCEPTION
    WHEN duplicate_column THEN RAISE NOTICE 'column "user" already exists in history type.';
END;

Changes
Include a brief summary of changes...

• added migrations table
• numbered sql file(s)
• added logic for applying migrations only if they are not present

Checklist

• Pull request is limited to a single purpose
• Code style/formatting is consistent
• All existing tests are passing
• Added new tests related to change
• No unnecessary whitespace changes