Implement picking rather than custom maths for mouse collision

[jacomyal]

Using picking for mouse collision would bring a lot of good things in sigma:

• Nodes / edges clicking would work for every nodes and edges programs, including curved edges, rectangle nodes...
• Checking what's below the mouse would be O(1), whatever the size of the graph

The cost would be:

• It will take some time to implement
• It will make writing program shaders a bit worse
• It will have a RAM cost:
  • We need to store an image the size of the container, where pixel colors represent top item index
  • We need to add to every programs the index of the item in addition to the color (but this will cost way less than PR Instanced rendering #1377 will make us win)

Here is one plan to implement this feature:

• Shaders have macros to be built either for RENDERING_MODE or PICKING_MODE. Rendering mode is as it is today, picking mode will feed every pixel with either transparent or the index of the item (node or edge) to display as the color.
• Each program takes a new argument, pickingMode. When true, the program will compile the shaders with PICKING_MODE.
• The rendering loop has a new canvas and WebGL context, not visible on screen, where it draws every edge then every node, using the "picking" programs.
• Both normal programs and picking programs use the same ByteArrays.
• The methods getNodeAtPosition and getEdgeAtPosition look at the color of the pixel at the position, translates it to the related item index if any, and returns the related ID.
• Migrate all programs

Any opinion on that @Yomguithereal?

[jacomyal]

Oh, and it might allow us to completely remove the quad tree.

[jacomyal]

We just merged a working version of the picking into the v3 branch.