As of now the only upkeep task is a cleanup of the databases by
running a vacuum and/or a wal checkpoint to reduce their sizes.

* Allow to open in memory database only in test.
* Avoid getting the same path from two different ways, with the only
  difference being the creation of the dir if it doesn't exist

And when the registration round is closed.

To not block the tokio runtime since tasks such as the vacuum can take
some time.

By checking that the tasks are done using the logs. The actual check
that the tasks do what they claims is done in the `SqliteCleaner` tests.

To make it use the same conventions as the aggregator.

* Rename signer `database::test_utils` `database::test_helper` and move
  it to a dedicated file.
* Copy the same `TestLogger`
* Sync aggregator & signer `CardanoTransactionImporter`

Like the aggregator for now the only upkeep task is to cleanup
the databases by running a vacuum and/or a wal checkpoint
to reduce their sizes.

Executed at the start of the `unregistered > registered` state machine
transition.

Note: In the integration test the store have been changed from
independent `MemoryAdapter` (that use hashmaps) to a `SQLiteAdapter`
with a shared in memory connection to make it closer to the production
environment.

* Since the UpkeepService now took this cleanup task.
* Remove the `Vacuum` option from the ConnectionBuilder as it was the
  only use.
* Remove `vacuum_database` function and instead use `SqliteCleaner`
  everywhere to streamline usage.

Since that would probably mean that some writing are currently occuring
on the database (like if the cardano transactions preloader is running).

* Make signer run upkeep after registration
* Make upkeep a separate task in aggregator runner
* Make `Configuration::get_sqlite_dir` always create the directory if it
  doesn't exist instead of only in Production environment. As this
  behavior doesn't make tests crash anymore now.
* Typo fix

This method will replace replace most usage of the
`reply::internal_server_error` as it analyse as it can change the http
error code returned based on the error given:
* by default the error is `500 - InternalServerError`
* For Sqlite Busy errors or RegistrationRoundNotYetOpened error the
  error code change to `503 - ServiceUnavailable`

Mostly replacing usage of `reply::internal_server_error` as the new
function can return varius status code depending on the error.

As it's used as the reply message for multiple server side errors, not
just `500 - InternalServerError`.

* Mithril-aggregator from `0.5.32` to `0.5.33`
* Mithril-signer from `0.2.155` to `0.2.156`
* Mithril-common from `0.4.24` to `0.4.25`
* Mithril-persistence from `0.2.13` to `0.2.14`