Pasting entities aren't always snapped to grid

[bstordeur]

Not sure when this started, but for a few versions now pasting some entities doesn't align them to the grid as it did before.
Here's an example, I copy and paste an monster_enforcer in latest TB, notice how its bounding box is not aligned properly to grid when I paste :

Now if open some older version of TB (2019.4 here, so not that old), if I do the same thing, it snaps as it always has :

System Information

TrenchBroom 2019.5 Build v2019.5-RC5-1-gf002ad8 Release on Windows 7.

Expected Behavior

Like it was before, snap the bounding box of the entity to the current grid setting.

Steps to Reproduce

Create a monster_enforcer (or most other entities), copy it, and paste it, notice how it's no longer aligned to the grid.

[kduske]

Probably using the wrong bounds for alignment. Are the gifs switched?

[bstordeur]

The gifs were switched when I first posted the issue, changed it right away though.
It looks like it only happens with ents that have a model, almost like it's using the models bounding box (which you can see when you hover over them).

[ericwa]

The bounds displayed when you mouseover a monster changed too, now it includes the model.
We should revert that to the old behaviour.

[ericwa]

@kduske There are different ways to approach this but somehow we need to separate "culling bounds" from the "semantic bounds" in Node's API.

What I'm thinking is Node::bounds() should be the semantically important bounds - so for point entities, this is the Entity::definitionBounds(), and we should add a new Node::cullingBounds() that is the maximum bounds including arms poking out, etc., used for rejecting hits when doing hit detection, frustum culling.

For this particular issue, the paste code is using MapDocument::selectionBounds() to place the pasted objects, so that needs to use the semantic bounds.

After using it a bit I'm not sure if there's really a point to drawing the outer box. Or if we do keep it, the text overlay should be describing the inner bounds, anyway.