Fixed missing scaling for MouseWheel events on Windows

[emuell]

Also, the PhysicalToLogicalPoint hack was missing for MouseWheel event. I guess that the same applies here, but I can't test that, because PhysicalToLogicalPoint and co don't do anything on my Windows setups.

Further, errors from ScreenToClient got handled with an assert! for MouseWheel events, but got silently ignored in the drag and drop impl. I've unified both, too, and do log them now. I think a warning is fine here, a panic would be too much, but that's your call to make.

NB: This is a regression of: dd350f8

[ilmai]

Just had a chance to test this. Looks fine to me, except for some reason mouse coordinates work just fine without the DPI-scaling hack. I'm a bit baffled honestly as to why ScreenToClient works differently when doing drag & drop and when handling mouse wheel events. I event did some logging and checked that both screen coordinates are the same for the upper left and bottom right corners of the window, so it's really just the conversion that's working differently for some reason. I think I'll look into this mystery a bit more before accepting this PR.

[ilmai]

Actually for now, please just remove the DPI hack code from mouse wheel event handling. I'll add a more descriptive comment for the drag & drop DPI hack and make a note of looking into this in the future.