Create a Base ReplayProcessor Class

The example from: `bin/replay_actions.py` shows a `ReplayProcessor` class that saves the stats from replay observations. This idea is extensible.

For running precise analyses on the in-game observations they may need to be saved into some other data structures (maybe even pushed into a database).

There is a need to create a more generic class to be able to override the `process_replay()` method.
This method steps through the replay and adds some statistics to `self`.

The idea is simple but there is multiprocessing involved and queues. The new approach should retain this functionality but allow for more customizability of the observation processing function.

Start by reading bin/replay_actions.py, especially the ReplayProcessor class and its process_replay() method, then trace how multiprocessing and queues support replay processing. Done means a generic base class preserves that functionality while allowing subclasses to customize observation processing and storage.

Written by the indexing model from the issue text.