Improve resampling/evaluation; add measures API (incl. LossFunctions)

[ablaom]

• Introduction of traits for measures (loss functions, etc):
  • allowing user to use loss functions from LossFunctions.jl,
  • enabling improved measure checks and error message reporting with measures
  • allowing evaluate! to report per-observation measures when available (for later use by Bayesian optimisers, for example)
  • allowing support for sample-weighted measures playing nicely with rest of API

The new traits are defined at the top of /src/measures.jl

• Improvements to resampling:

  • evaluate! method now reports per-observation measures when available
  • sample weights can be passed to evaluate! for use by measures that support weights
  • user can pass a list of train/evaluation pairs of row indices directly to evaluate!, in place of a ResamplingStrategy object
  • implementing new ResamplingStrategies now straightforward
  • one can call evaluate (no exclamation mark) directly on model + data without first constructing a machine, if desired

• Substantial updates to the documentation to cover the changes. See, in particular new sections "Evaluating model performance" and "Measures".

Commits before rebase:

add evaluate! for user-specified resampling; needs integration with tuning

refactor cv to be special case of general evaluate!

integrate evaluation changes with tuning

update docs

add support for weighted measures in evaluuate!; tuning needs tests

update cheatsheet

add show method for Measure objects

Fix weights in resampler/tuning and add tests

Implement an MLJ interface for LossFunctions pkg

reports_each_sample->reports_each_observation per_sample->per_observation

fix LossFunctions MarginLoss's to be probabilistic

Add pretty printing to evaluation!

update docs

update docs

add measure/model compatibility checks

make resampling interface more customizable

fix @test_logs

update docs

update docs

update doc string

improve measure checks

update docs