Mention `tape_copy` in doc or expose a deepcopy flag ?

[FredericWantiez]

From what I understand Libtask does not deepcopy struct by default when used in TapedTask. For example this shares the initial model:

mutable struct Model
    t::TArray
    n::Int
    Model(n::Int) = new(TArray(Float64,n), n)
end

function (model::Model)()
    model.t[1] = 1
    Libtask.produce(model.t[1])
    for i in 2:model.n
        r = rand(Normal())
        model.t[i] = model.t[i-1] + r
        Libtask.produce(model.t[i])
    end
end

model = Model(4)
ttask = Libtask.TapedTask(model)

Libtask.consume(ttask) # -> 1

ttask2 = copy(ttask)

Libtask.consume(ttask)   # -> 1 + r1
Libtask.consume(ttask2) # -> 1 + r1 + r2 (!)

To get the expected result we need to add:

Libtask.tape_copy(model::Model) = deepcopy(model)

which is hidden somewhere (the comment itself is very helpful)

Libtask.jl/src/tapedfunction.jl

Lines 351 to 360 in f4941a0

	"""
	tape_copy(x)
	Function `tape_copy` is used to copy data while copying a TapedTask, the
	default behavior is: 1. for `Array` and `Dict`, we do `deepcopy`; 2. for
	other data types, we do not copy and share the data between tasks, i.e.,
	`tape_copy(x) = x`. If one wants some kinds of data to be copied, or
	deeply copied, one can add a method to this function.
	"""
	function tape_copy end

Maybe we could mention it somewhere in the readme since it's required to get AdvancedPS.PG running properly.