Lightweight benchmarking on Github Actions

[yebai]

We would like to gradually move some simple benchmarking models into Turing.jl/benchmarks/* from TuringBenchmarks. These models would form a lightweight benchmarking suite, which we can run on Travis for every new PR. After Travis completion, the benchmarking results should be posted back as a comment to the PR. This allows us to have a quick inspection before merging PRs.

This is already partially implemented by #816, although we still need to add some models to Turing.jl/benchmarks/ and write a runbenchmarks.jl script.

Example models

• gdemo
• MvNormal
• Bayesian Logistic Regresion

and

• comparison of different Turing samplers see e.g. https://dm13450.github.io/2019/04/10/Turing-Sampling-Speed.html

Any comments?

[mohamed82008]

I like the idea!

[yebai]

Closed in favour of TuringLang/DynamicPPL.jl#346