Our old (base-path -> data-path) SQLite migration can't work otherwise.

It's probably not necessary to keep it anymore, but since we still do,
at least we should take care to ensure it works.

Should have been part of 149872e00c

This also adds support for installing python3-docker (not python-docker)
in systems that run Python 3.

Raspbian doesn't seem to support arm64, so this is somewhat pointless
right now.

However, they might in the future. Doing this should also unify us
some more with `setup_debian.yml` with the ultimate goal of
eliminating `setup_raspbian.yml`.

It's likely installed by default, but it doesn't hurt to specify it.
It also makes us more the same with `setup_debian.yml`.

Well, `ubuntu` or `debian`, the same key is served right now,
so it doesn't really matter.

This seems cleaner and less prone to breakage though.

Until now, we've only supported non-amd64 on Raspbian.

Seems like there are now people running Debian/Ubuntu on ARM,
so we were forcing them into amd64 Docker packages.

I've gotten a report that this change fixes support
for Ubuntu Server 20.04 on RPi 4B.

We were only defining this in `group_vars/matrix_servers`, which is
inconsistent with how we normally do things.

Revert "Correct inabillity for appservice-discord to connect"
This reverts commit 673e19f830.

While certain things do work even with such a local URL, sending
messages leads to an error like this:

> [DiscordBot] verbose: DiscordAPIError: Invalid Form Body
> avatar_url: Not a well formed URL.

Fixes https://github.com/Half-Shot/matrix-appservice-discord/issues/649

The sample configuration file for appservice-discord
c29cfc72f5/config/config.sample.yaml (L8)
explicitly says that we need a public URL.

Now that 0.7.2 is out, the Docker image supports Postgres
and we can do the (SQLite -> Postgres) migration.

I've also found out that we needed to fix up the `tokens.ex_date` column
data type a bit to prevent matrix-registration from raising exceptions
when comparing `datetime.now()` with `ex_date` coming from the database.

Example:

> File "/usr/local/lib/python3.8/site-packages/matrix_registration/tokens.py", line 58, in valid
> expired = self.ex_date < datetime.now()
> TypeError: can't compare offset-naive and offset-aware datetimes

In cases where pgloader is not enough and we need to do some additional
migration work after it, we can now use
`additional_psql_statements_list` and
`additional_psql_statements_db_name`.

This is to be used when migrating `matrix-registration`'s data at the
very least.

This switches us to a container image maintained by the
matrix-registration developer.

0.7.2 also supports a `base_url` configuration option we can use to
make it easier to reverse-proxy at a different base URL.

We still keep some workarounds, because of this issue:
https://github.com/ZerataX/matrix-registration/issues/47

We were running into conflicts, because having initialized
the roles (users) and databases, trying to import leads to
errors (role XXX already exists, etc.).

We were previously ignoring the Synapse database (`homeserver`)
when upgrading/importing, because that one gets created by default
whenever the container starts.

For our additional databases, it's a similar situation now.
It's not created by default as soon as Postgres starts with an empty
database, but rather we create it as part of running the playbook.

So we either need to skip those role/database creation statements
while upgrading/importing, or to avoid creating the additional database
and rely on the import for that. I've gone for the former, because
it's already similar to what we were doing and it's simpler
(it lets `setup_postgres.yml` be the same in all scenarios).

We were referencing them from `group_vars/matrix_servers` since
recently, but there were no such variables and they weren't being put to
use.

Auto-migration and everything seems to work. It's just that
matrix-registration cannot load the Python modules required
for talking to a Postgres database.

Tracked here: https://github.com/ZerataX/matrix-registration/issues/44

Until this gets fixed, we'll continue default to 'sqlite'.