Potential breaking change for brms 3.0: no special treatment for mu

[venpopov]

This is a summary of our recent discussions in person

Overview

General idea is to move away from treating the mu parameter in any special way. This would involve:

• Do not require models to have a mu parameter
• No distinction between mu and distributional parameters
• For models that do have a mu parameter, do not redact mu anywhere (e.g., Intercept_mu, not Intercept, do not hide in priors, etc)

This would be a breaking change, so delegated to v3.0.

Additional context

Here are some relevant previous issues/PR requests. I have closed them as we discussed, and the current issue consolidates the info in one place.

#1610
#1596