Add empirical HMC

[rlouf]

https://arxiv.org/pdf/1810.04449.pdf

This requires:

1. A jittered version of HMC (as in Add new CheesHMC sampler by Hoffman et al #12);
2. A new tuning algorithm that returns an array of trajectory length;
3. Dual Averaging;
4. Mass matrix adaptation;
5. The NUTS kernel to use with (1).

(1) is trivial to
implement with BlackJAX and (3,4,5) are part of the NUTS sampler.

[sethaxen]

Some more discussion and comparisons here: https://discourse.mc-stan.org/t/hmc-jittered-vs-nuts-on-1000-dimensional-standard-normal/8383

[rlouf]

The Stan Discourse will never cease to amaze me.