add datatype to OneHotEncoder and ContinuousEncoder

[tiemvanderdeure]

allows users to specify what datatype OneHotEncoder and ContinuousEncoder return, so that this now works:

using MLJModels, MLJBase
X = (x = rand(10), c = categorical(rand(Bool, 10)))
mach = machine(ContinuousEncoder(datatype = Float32), X)
fit!(mach)
transform(mach, X)

see discussion here: #565

There is probably a way to fix this for ContinousEncoder without copying data around just as much. I'll figure that out if we decide to implement this.