Add Yogi optimizer description, reference, arg doc strings to Optax

[8bitmp3]

This PR adds a short description, a paper reference, and arg doc strings to the Yogi optimizer in Optax, since it was not documented. Related to #71 by @mtthss

  ...
  Yogi is an adaptive optimiser, which provides control in tuning the effective
  learning rate to prevent it from increasing. By doing so, it focuses on
  addressing the issues of convergence and generalisation in exponential moving
  average-based adaptive methods (such as Adam and RMSprop). Yogi is a
  modification of Adam and uses the same parameters.
  ...

Used Google Scholar/Research to obtain a short link for the paper: http://www.sanjivk.com/yogi_nips2018.pdf. The alternative is https://papers.nips.cc/paper/2018/file/90365351ccc7437a1309dc64e4db32a3-Paper.pdf but it may not pass the lint test @mtthss.