core: Respect ratio field for non-morph shapes (fix #1291)

[Herschel]

This changes the logic for rewind-gotos to respect the ratio field inside the PlaceObject tags. See #1291 for a full description.

Previously we would manually keep track of the placement frame for a display object whenever it was instantiated. A rewind would immediately remove all objects that were created after the target frame (because they should not exist on that frame). However, now we need to inspect the ratio field of the final PlaceObject tag to decide whether to remove an object; so now we defer the decision of whether to remove an object until after we churn through all the frames.

Still needs some more research+testing to verify this doesn't regress anything. I need to add a test case to verify the new behavior. My initial tests indicate this does fix #1060, which makes sense because this SWF is made by a 3rd party tool.

[Toad06]

There are some issues where things remain on screen while they shouldn't:

• In DTunnel (https://www.newgrounds.com/portal/view/161144), click "Instructions", then "Back". The instructions will remain on screen.

• In Honey Bunny (https://www.newgrounds.com/portal/view/182750), start a game. Press Up and get the yellow key, then go down and enter the "Sweet home". Click "Give honey", the button doesn't disappear while it should; click "Leave cottage", the game will resume but what is drawn on screen will not change, so the game becomes unplayable.

• In Luigi's Day (https://www.newgrounds.com/portal/view/44505), start the game and fall in the water. The audio should stop (but doesn't on the Web backend only - that's a regression but doesn't seem to be related to this PR?). Click "Replay", the button will remain on screen.

• In Ratman Ralph (https://www.newgrounds.com/portal/view/67006), when you click "Options" or "Instructions", then "Next" (at least one time) and "Close", the text remains on screen. However, the PR fixes the last issue described in [Ratman Ralph] Differences with the Flash version #815.

Other issues:

• In Zed (https://www.newgrounds.com/portal/view/89118), transitions between rooms are broken (screen should be filled with blue squares but only one is created on the top left corner).

• In Ultimate Sonic Flash (https://cdn.discordapp.com/attachments/610531541889581068/752810022915801128/sonic_ultimate.swf - edited version to bypass the loading screen), start a new main game with Sonic. Move the character and you will notice that platforms randomly appear/disappear which completely mess up the game.

[Herschel]

Thx for testing! Could you try again with the latest changes?

[Toad06]

Hmm, I'm still observing the same issues (is it because the PR seems to miss a piece of code from commit 04e3f1ecea9572801a92ac6596134fa5a174b5ba? I'm not seeing the if is_rewind { code in "movie_clip.rs").

[Herschel]

Whoops, you're right, I futzed up a rebase. Should be okay now.

[Toad06]

Almost all of the games above work as intended now! 👍 Only Zed still has issues with transitions (for the audio stuff on Luigi's Day, I will open a bug report).

I've also tested other files, unfortunately it turns out some of the games by Ferry Halim have issues:

Starry Night (http://www.ferryhalim.com/orisinal/g3/starry.htm)
No stars to light up appear.

Critical Zone (http://www.ferryhalim.com/orisinal/g2/criticalzone.htm)
The game no longer works.

The Bottom of the Sea (http://www.ferryhalim.com/orisinal/g3/sea.htm)
The more the character gets down in the water, the more the background gets truncated (or doesn't repeat itself).

[Herschel]

Thx for the testing!

Upon thinking about this some more, I think better would be to change GotoPlaceObject to be an enum with a Remove variant, and keep track of removes as well as places. For a rewind, we can initialize the goto command list with a remove for every current display object (because a rewind starts from an empty frame 1). This removes the need for the awkward "GotoPlacedObject" flag. I also think this will be more accurate for the execution order of clip unloads (we haven't really paid attention to this yet).

[Herschel]

Closing in favor of #5431.