Job manager: Move jobdatabase classes out of __init__

[VictorVerhaert]

the init file currently contains a lot of logic making it hard for new users to figure out how it works.

One way of cleaning this file is moving job database logic to a separate file `job_database.py' (and including the stac api one)

In order to prevent existing links from breaking we can import the databases and add them to the all variable as described here:
https://docs.astral.sh/ruff/rules/unused-import/
(I realize we don't use ruff as a linting tool but still believe we should add it to all)

[soxofaan]

fully agree to move all implementation out of job_management/__init__.py into submodules

However, there are a couple of ongoing job manager PRs, so we have to be careful when to start and merge the move to avoid branch conflicts

[HansVRP]

one issue I forsee is a potential circular import;

FullDataFrameJobDatabase depends on the multibackend job manager's _normalize_df

[soxofaan]

yes there is a circular dependency there at the moment. But I think it can be broken by moving _COLUMN_REQUIREMENTS and _normalize_df out of MultiBackendJobManager. Neither of these are directly used by MultiBackendJobManager anyway (except if you look closer for hidden coupling).