An immutable type ProximalPoint about proximaloperators.jl HOT 1 OPEN

I think this is a very interesting idea, but I think we have to think it through really carefully. I think there are a couple of trade-offs regarding how much data should be saved to be efficient on multiple calls versus being efficient on a single call.
Do you think this should be the output even for prox!? This could potentially lead to a lot of extra allocations, for example in

for i = 1:100000
   prox!(y,f,x)
   x = 0.5*x .+ 0.5*y
end

we would allocate one point::AbstractArray in every call. Unless we let this element refer to the vector y, in which case the user might inadvertently change the elements inside the ProximalPoint. (Immutable only means that it will reference the same array, the elements in the array might still change).

As a side-note: We should probably not let the definition be

immutable ProximalPoint <: AbstractArray
  f::ProximableFunction
  point::AbstractArray
  value::Real
end

but rather

immutable ProximalPoint{F<:ProximableFunction, P<:AbstractArray, V<:Real}
  f::F
  point::P
  value::V
end

I created a related issue #23 for the existing problems of this type.

from proximaloperators.jl.