Taking inspiration from Flyway, this commits adds the undo subcommand to
the application, with the `--step <n>` flag.

Undo scripts will have the `U` prefix, with the same name and version as
their corresponding versioned script. It will attempt to undo <n> amount
of versioned scripts until it finishes, or no corresponding undo script
is found.

Under the hood, the application is removing the `V` script from the
change history table, so when the `deploy` command is ran, the versioned
script can be applied again.

Algorithm:

* Fetch all applied `V` migrations
* Fetch all `U` migrations
* Get `U` migration of last `V` migration.
* IF `V` migration has no `U` migration, process stops
* Apply `U` script
* Delete `V` script from the change_history_table related to `U`

This commit closes issue Snowflake-Labs#19

[Swithing to Debian 12](salemove/base-images#46) in python base image
caused [OpenSSL switch](salemove/base-images#46 (comment)) from 1.1.1 to 3.0.

Current version of `snowflake-connector-python` 2.8.0 which depends on
`pyopenssl` 22.1.0 fails because of OpenSSL switch https://ci.at.samo.io/job/salemove/job/snowflake-ingestion/job/PR-589/8/consoleFull

After the update of `snowflake-connector-python` to 3.7.0 `pyopenssl`
becomes `23.3.0` and doesn't fail anymore in this image.

Tested `snowflake-ingestion` in [this PR](salemove/snowflake-ingestion#600).
[Pipeline](https://ci.at.samo.io/job/salemove/job/snowflake-ingestion/job/PR-600/6/consoleFull)
doesn't fail anymore, and does migration/rollback successfully.

CBRO-1908

CBRO-1908

In case you clone a database, you would want to recaclulate checksums for
repeatable scripts as they already exist but checksum is wrong as database name
has changed.

Both deploy_command and recalculate_checksum use very similar body. Only
differences are:

1) Deploy deals with V, R, A migrations, while recalculation deals only with
R migrations.
2) Deploy also applies the script, while recalculation on updates
schemachangetables.