Add new traits to to support better MLJ machine type checks and composite models

[ablaom]

Add the following traits with specified fallbacks:

<> fit_data_scitype = Unknown - for the scitype of data in fit(model, verbosity, data...) calls (and whence machine(model, data...) calls). For example, a supervised classification model, the value might be something like like Tuple{Table(Continuous),AbstractVector{Finite}}. If the model additionally supported weights, the value might be

Union{Tuple{Table(Continuous),AbstractVector{Finite}},
      Tuple{Table(Continuous),AbstractVector{Finite},AbstractVector{Continuous}}

For current abstract Model subtypes, this trait would have a fallback inferred from existing traits, but this allows for models to have one or more custom fit signatures, and allows for a better fallback check for data bound to machines.

<> predict_scitype = Unknown - a guaranteed upper bound for the scitype of the output of predict method, as distinguished from target_scitype which is the scitype of the training "target". For supervised models these only agree in the Deterministic case, not the Probablistic case, which generally involves Density or Sampleable scitypes. For example, a probabilistic classifier might get the value AbstractVector{Density{<:Finite}}. For existing Model subtypes, this is to be inferred as far as possible from the target_scitype and prediction_type traits.

<> transform_scitype = Unknown, inverse_transform_scitype=Unknown - similar to above predict_scitype.

<> abstract_type = Any - this will generally be the direct supertype of the model. So, for example, a probabilistic classifier will get the value Probabilistic.

[codecov-commenter]

Codecov Report

Merging #18 (5e6426f) into dev (d70ee0c) will not change coverage.
The diff coverage is 100.00%.

@@            Coverage Diff            @@
##               dev       #18   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files            1         1           
  Lines           54        59    +5     
=========================================
+ Hits            54        59    +5     

Impacted Files	Coverage Δ
src/StatisticalTraits.jl	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d70ee0c...5e6426f. Read the comment docs.