For that project I was able to put SQL DDL+DML+stored procedures in version control, create/run stored procedure (TDD even) unit/integration tests on mock data against other stored procedures, had pass-fail testing/deployment in my CICD tool right alongside native app code, and did some rollback support (although that was more trouble than I think it was worth), all using Liquibase change sets (+git+Jenkins). Flyway could also have worked but we were using Liquibase.
It did require some creative thinking to apply changesets and preconditions and post conditions and ability to have stored procedures execute and read results of other stored procedures.
Last I checked, the system had been used over many years to process/evaluate $50 billion in order transactions.
One major benefit of stored procedures, in addition to separation of concerns, is that you can declare them SECURITY DEFINER and give them access to tables the Python process doesn't (in a way reminiscent of setuid), thus improving the security posture dramatically.
One example: you could have a procedure authenticate(login, password) that has access to the table of users and (hashed) passwords, but if the Python app server is compromised it doesn't have access to even the hashed passwords or even the ability to enumerate users of the system.