AdvancedVI 0.3 and compatibility with Turing.jl

[mhauru]

The new DynamicPPL.LogDensityFunction interface paves the way for this...

[Red-Portal]

Hi all! I added a PR for this: #2506 , which is already functional. Things weren't as a big deal as I was worried. What remains is to converge to an interface for exposing the features of AdvancedVI, and this will also involve deciding how much breaking changes we are willing to allow.