Description

This PR wires up the River client into the JIMM domain layer, allowing us to insert river jobs as necessary.

The main challenge I faced with this was the potential for circular dependencies. To resolve this I've made a package called rivertypes, thanks @luci1900 for the discussion on this. This package holds the job arguments that are used when inserting a job using the river client. The package go doc explains this nicely,

// River job args are analogous to a schema for traditional queue systems, defining what
// we will insert into the DB for processing later by workers. By defining them
// in a 3rd package we avoid circular dependencies between the domain layer and the

There was one more issue with the tests, to try and summarise

• internal/river depends on types from internal/jimm/upgrade for its UpgradeTo interface definition (though technically the current methods only use base types, but I've added a blank import to simulate the future problem).
• That revealed an import cycle in tests. The river tests create a DB using internal/jimmtest/ helpers. The import cycle came from the fact that internal/jimmtest/ imports internal/jimm/ and internal/jimm/ imports the thing river wrapper we have to make it easier to mock in tests.
• The solution was simply to move the DB related helpers into internal/jimmtest/dbtest/

I recommend reviewing the commits separately.

Fixes JUJU-9057

Engineering checklist

• Documentation updated
• Covered by unit tests
• Covered by integration tests