Rework database session handling

See merge request aura/engine!11

Currently the database session is created during application startup and
is used in all threads.  This is problematic, since the session is not
thread-safe.
Now we use the request-based database session provided by
flask-sqlachemy instead.

But this does not really solve the problem of using the same session in
different threads, since the threads are started within a request.

The introduction of short-lived sessions caused issues with functions,
which return database objects.  Such objects are not accessible (due to
lazy loading), after the session is closed.

Instead we should either move the session scope the respective callers
or use a request-based session handling (e.g. the one provided by
flask-sqlalchemy).

This reverts commit 0b594e50 and 70ffd1a0.

See: #93

refactor: use short-lived sessions when accessing the database

Closes #75

See merge request aura/engine!9

feat(database): add support for sqlite database

See merge request aura/engine!5

docs: use local Debian packages for bare-metal deployment

See merge request aura/engine!6

# Conflicts:
#   docs/bare-metal-installation.md

Grammar and other documentation issues

See merge request aura/engine!7

# Conflicts:
#   src/engine.py
#   src/mixer.py

fix(heartbeat-monitor): port the "PyHeartBeat" script to Python3

See merge request aura/engine!8

docs: fix spelling issues

See merge request aura/engine!4

Improve graceful shutdown. #88

See merge request aura/engine!10

Thanks, codespell!

The Debian packages benefit from security support and should be much
easier to install for most users.

Currently all requirements of Auro Engine are packaged in Debian.
Thus the virtualenv-related documentation snippets can be removed.

An sqlite database is much easier to setup for simple local deployments
and specifically for test automation.

Previously the database connection ("DB.session") was established during
the startup of Aura Engine.
This was causing conflicts in the threaded execution environment (#75).
Now all sessions are acquired in a short-lived context:

  with DB.Session() as session:
      ...

If database query performance is a real concern, then the session
contexts could be moved to a higher level.

Closes: #75

In addition to porting it to python3, a number of documentation and code
issues were solved:

* doc: proper order of arguments
* doc: mention debug mode (previously via "__debug__", now "DEBUG=1")
* doc: provide example for manual heartbeat packets
* code: specify timeout for "socket.recvfrom" (avoid hanging)
* code: handle parameters properly (cast to int/float)
* style: format with "black"