implement M3 model

[GidonFrischkorn]

@venpopov: Just as information that I will be working on this together with Isabel & Chenyu. We are planning to implement the following things:

• Add out-of the box implementations of M3 for simple and complex span (potentially also for removal)
• Add M3 custom for allowing to specify custom activation equations and predict the activation sources by experimental variables
• Potentially support both Softmax & Luce's choice rules as decision rules (explore additional possibilities)
• Implement post-processing routines to make results more accessible (potentially contact Paul about contributing to brms)
• Chenyu will add more nice post-processing functions.
• Simulate data from M3 models

We have created a fork, where I will work with them independently.