Improve/standarize the form of `fitted_params` for transformers

[ablaom]

fitted_params(model, fitresult) should return a named tuple with informative keys. Ideally, the user should not need to look up documentation to grep the results. Here's an example of a poor implementation from this package:

model = Standardizer()
X, y = make_regression()
mach = machine(model, X) |> fit!

julia> fitted_params(mach)
Dict{Symbol, Tuple{Float64, Float64}} with 2 entries:
  :x2 => (0.14288, 0.858847)
  :x1 => (-0.0392576, 0.977905)

The complete list of bad transformers needing attention are:

• UnivariateDiscretizer
• Standardizer
• OneHotEncoder

The "Fitted parameters" section of each docstring will need to be updated at the same time.