Fix big markers

[griwodz]

With this change, I feel that it's easier to understand how layers map to collections, and I can find 4 markers in the large sample image that gave you trouble.

[griwodz]

Why yet another unique_ptr, this time inside a vector? Am I stupid?

Well, EdgePointCollection has no copy constructor, it is forbidden with =delete for a good reason: it contains lots of unique_ptrs to arrays, and copy constructors mess things up because they seem to create identical copies, while they are in reality invalidting the source object.
Forbidding the copy constructor means that vEdgePointCollections cannot be of type vector<> because vector<> leads to compile errors when the copy constructor of its type does not exist.
list<>, on the other hand, was the reason why we got into this mess in the first place. Since we cannot use operator[] on list<>, we (probably I) created this bug where edges were not stored in the right layer.

So I tried vector<EdgePointCollection*> first, and that gave the correct result for me as well. But the host code is full of exception raising and exception checking stuff, and that would probably lead to memory leaks.

vector<unique_ptr > is a bit uglier, but it calls delete on the EdgePointCollection objects when leaving the context, even in case of an exception.

[griwodz]

The other image, where a single large marker in the center was missing, works for me as well.

[fabiencastan]

Maybe we could use boost::ptr_vector which is done for this use case?