Algorithm Request: more DQN-based approaches

I am thinking of using sheeprl as the base for my RL experiments! My work usually builds off of DQN-type algorithms: in increasing level of complexity, off of DDQN, Rainbow, or R2D2. Having some of these implemented would make this library much more convenient for research IMO.

It's super cool that you have Dreamer and Plan2Explore implemented, but that's not a great starting point for RL research because of how complex and opinionated they are. It would be great to just have a simple DDQN baseline implemented! All the PPO and A2C stuff is great if you work in the online setting, but DQN type things would let us build things for batch training. In a perfect world it'd also have an implementation of something like R2D2 since that's a good SOTA-ish DQN upgrade.

Excited to start trying it out!

The request covers several possible additions, including DDQN, Rainbow, and R2D2, but names no files, tests, or entry points. First narrow the scope to a specific baseline and review the existing PPO, A2C, Dreamer, and Plan2Explore implementations; done should include a working DQN-based algorithm suitable for the library's batch-training research use case.

Written by the indexing model from the issue text.