Sync with Godot master

[lufog]

For fun, I'm trying to compile this module on the master branch.

Task list:

• Refactor module initialization godotengine/godot#60723
• Remove RES and REF typedefs in favor of spelled out Ref<> godotengine/godot#60714
• Replace most uses of Map by HashMap godotengine/godot#60999
• Rename StreamTexture* to CompressedTexture* godotengine/godot#58788
• Discern between VIRTUAL and ABSTRACT class bindings godotengine/godot#58972
• Rename Transform2D and Basis elements to columns and rows respectively godotengine/godot#60627
• Make FileAccess and DirAccess classes reference counted. godotengine/godot#59440
• Remove support for PVRTC texture encoding and decoding godotengine/godot#56309 (Idk what to do with it so commented it out to compile)

Btw, I'm not sure that I can do everything on my own. And Idk what I'm doing. Therefore, all interested don't hesitate to join.

[lufog]

Hooray! The module is compiles and works with Godot master branch. From unresolved:

• I have no idea what to do with PVRTC.
• Most likely I broke a lot of things during 49e0449, but I'm tired, and I'll deal with it later.

[nikitalita]

Thank you very much for this! I was really dreading updating the master sync for release, so I'm glad someone did it for me :)

I added a few commits:

• the file format extension for v4 textures changed, so I added those extensions to the relevant filters
• the removal of PVRTC support means that we can no longer open v2 or v3 textures in that format; It doesn't appear to have widely been used, so I just removed support.
  • Irritatingly, this means that the Image::Format enum changes between v3 and v4, so I had to create a compatibility function to convert v3 image format numbers to the correct enumeration value and
• FileAccess changing to a ref-counted object introduces unintended consequences. We have to make doubly sure that:
  • The fileaccess actually goes out of scope when we are done reading/writing. If the reference hangs around, that means the file will remain open forever, so we have to explicitly unreference it.
  • If we need to read a file right after writing it, we have to explicitly call flush() or make sure that the fileaccess object is actually unreferenced, or unwritten bytes will hang around in the buffer and not be written to disk
• Every time we sync with master, we have to make sure to change the sync reference commit in the CI scripts, or the CI builds will fail

It's surprising to me just how much the API is changing at this late in the development cycle; I probably won't update the sync again until v4 gets closer to release, I think.

[nikitalita]

CI is currently broken, but I've validated this locally. I'll PR a fix for the CI shortly.