Refactor overapproximate: two-stage dispatch

[schillic]

I suggest to let overapproximate only dispatch on the second argument (the target type). In each method we can call helper methods that then dispatch on the first argument. This unifies the interface and avoids many disambiguation methods.

function overapproximate(X::LazySet, ::Type{<:Interval})
    return overapproximate_interval(X)
end

function overapproximate(X::LazySet{N}, ::Type{<:HPolygon}) where {N<:Real}
    return overapproximate_hpolygon(X, N(1e-3))
end

function overapproximate(X::LazySet{N}, ::Type{<:HPolygon}, ε::Real) where {N<:Real}
    return overapproximate_hpolygon(X, ε)
end

...

function overapproximate_interval(X::LazySet)
    ... # default implementation with two support-function queries
end

function overapproximate_interval(X::S)
    ... # specific implementation for set type S
end

...

function overapproximate_hpolygon(X::LazySet{N}, ε::N) where {N<:Real}
    ... # epsilon-close approximation
end

...

Originally posted by @schillic in #1353 (comment)