[3.x] Make autotiles fall back to the most similar bitmask

[wareya]

Implements godotengine/godot-proposals#3851 for 3.x. If the desired bitmask doesn't have any hits in the tilemap, it searches the tilemap for tiles with the most similar bitmask and uses that instead.

Not needed for 4.x because the terrains system already behaves similarly.

Test project in this comment: #71533 (comment)

I need feedback:

1. What is the standard way to define helper functions in the Godot codebase, like my _score_bitmask_difference and _count_bitmask_bits? Should they belong to a class or are they okay standing in open scope in a cpp file?
2. Should this be behind a toggle? It's a compatibility break to anyone generating maps from code with broken/incomplete autotiles.
3. If I put it behind a toggle, should it be on TileMap or somewhere else?
4. What pieces of documentation do I need to update?

[wareya]

Updated to latest 3.x, added an option (on by default) to TileSet, and updated the docs. The autotile tutorial is slightly out of date, but it's in another repo and I'm not sure what the process is for updating tutorials.

I'm still not sure whether it's OK to leave helper functions like _score_bitmask_difference and _count_bitmask_bits floating around in global scope, and I'm not sure I hooked everything up in the editor properly (the tileset editor plugin has an extra layer of abstraction around tileset options), so it would be nice if someone went over this.

[KoBeWi]

What is the standard way to define helper functions in the Godot codebase, like my _score_bitmask_difference and _count_bitmask_bits? Should they belong to a class or are they okay standing in open scope in a cpp file?

They are normally part of a class.

[wareya]

Thanks. Rebased and made the helper functions static class functions.

[KoBeWi]

Do you maybe have some example TileSet (texture + resource) to make testing this easier?

[wareya]

Test project:
TileTest.zip

[KoBeWi]

I tested the feature and it does visibly improve autotiling. I'm not familiar with the implemented algorithm, but the code is clean and works™, so the implementation is alright I guess.

As for the concerns about compatibility breakage mentioned in the proposal, this only affects newly drawn tiles in the editor, no? Then I don't think a different default behavior is a problem.
(although maybe it does affect procedurally generated TileMaps; not sure if it's fine 🤔)

[akien-mga]

Thanks!