Add `eugene ci` command

[kaaveland]

High level behaviour: eugene traces all new migration scripts, then posts migration reports to the pull request / merge request.

There is quite a bit to unpack here, and lots of problems to solve. How can we minimize the amount of setup and configuration necessary, so developers will want to use the tool?

Here's a non-exhaustive list of problems to solve and some ideas:

• Discover new migration scripts. It is probably OK to say that we support only git for the start, so we could use something like https://github.com/rust-lang/git2-rs. We should probably require the developer to point eugene at some subtree instead of looking for all .sql files. We're probably going to need a git ref that we're making a pull request against.
• Use the migration tool for the repo to run all migrations prior to the new migrations, in the right order. We could probably ask the user to provide a command we can use to run all migrations, then we could temporarily move the new ones out of the way and move them back one by one, running the migration command in between.
• Figure out the correct order to run new migrations in, in case there are several. Not sure how to do that in the general case, but if we can discover naming standards with versions in them, we may be able to. Liquibase does not need versioned file names. One idea may be to ask the user to provide a shell command for sorting the new migrations in a config?
• One option is to create a database, apply the migrations to it, then use the flyway or liquibase tables to find the correct migration order, drop it, then do the migration again to inject traces in-between the migrations. This would require us to recognize the migration tool (or to explicitly support it). Maybe that's the right way?
• We should probably add some special support for refinery if we're adding special support for flyway and liquibase, to be a good citizen.
• Is it an option to just require the files to sort in correct order when sorted lexicographically, then eugene can just run them all with --commit without even invoking a migration tool? Or maybe we can require that each file contains a comment with a sequence number or something?

I am slightly worried about the prospects of doing this without teaching eugene lots of information about how flyway and liquibase work -- those are the two tools I meet at work the most often, and where I could most easily find users.

[kaaveland]

Just realized that #18 actually means there are migrations that can be run only once and can't be rolled back, which means we probably must apply them with eugene to generate the reports in CI, instead of using a migration command. Copying databases doesn't help us either, since stuff like create role is global to the server. 😭

If we want to get really dirty, we can use pg_basebackup to work around this, but that seems very involved.

[kaaveland]

With the new --git-diff flag, I feel like CI is a solved problem and what we need to do here is to set up some examples (one for github actions and one for gitlab ci, for now).

[kaaveland]

I feel like this is done now, going to tackle setting up some examples in https://github.com/kaaveland/eugene/issues/26 after I've released.