Fix unexpected control position change when left/top offsets not match pos_cache (reverted)

[L2750558108]

In #104357, in order to prevent the size of the offsets from being broken during offsets recalculate in set_position, the rect which use size calculated in real time via the offsets (previously use size_cache) is used in the offsets' recalculating

But this doesn't take into account the position of the rect. If the current pos_cache does not match (edge[0], edge[1]) calculated via offsets (e.g. the control's size is propped up by the minimum size when the grow_direction is not Right/Bottom), the position of the control may be offset to the top left of the expected position. This pr solve this

Fixes : #105771

[KoBeWi]

It fixes the regressions, but it might introduce a new one, just like the previous fixes.
Ideally we should have tests at least for each issue in the chain.

[KoBeWi]

Well, I verified that it does not bring back any of the related issues. Here are reproduction scenes for each of them:
Sizebugs.zip
(I did not test each one in their bugged state, only with this PR)

While there is a risk this PR will create just another regression, the alternative is reverting all previous fixes (or picking the least bad of the issues). Merging the fix while we are still in dev stage is fine; potential new bugs are likely to get detected and fixed before stable (and assuming we don't re-introduce old bugs, there is only a limited number of things that can break, right? 🤞).

[akien-mga]

YOLO I guess :D

We really need to work on a test suite for Control positions as you suggested, as we've had probably a dozen such regressions over time already.

But yeah, let's give this one a spin, and if we found more regressions which require piling more hacks on top, there's always the option of reverting the whole chain of PRs to the last "reliable with known issues" state.

[Repiteo]

Thanks!