Fix compiling on arm64 Linux with GCC

[aaronfranke]

I tried to test compiling something in my arm64 Linux VM, specifically Ubuntu 24.04.2 LTS with GCC 13.3.0 and kernel 6.8.0, and I ran into this error:

[ 16%] Compiling modules/gltf/gltf_document.cpp ...
[ 99%] In file included from ./core/templates/vector.h:44,
                 from ./core/templates/local_vector.h:37,
                 from ./core/object/message_queue.h:36,
                 from ./core/object/object.h:35,
                 from ./core/variant/binder_common.h:35,
                 from ./core/object/method_bind.h:34,
                 from ./core/object/class_db.h:34,
                 from ./core/object/ref_counted.h:34,
                 from ./core/io/resource_uid.h:34,
                 from ./core/io/resource.h:34,
                 from modules/gltf/extensions/../extensions/gltf_light.h:34,
                 from modules/gltf/extensions/../gltf_state.h:34,
                 from modules/gltf/extensions/gltf_document_extension.h:34,
                 from modules/gltf/gltf_document.h:34,
                 from modules/gltf/gltf_document.cpp:31:
In member function 'CowData<T>::Size CowData<T>::size() const [with T = signed char]',
    inlined from 'Vector<T>::Size Vector<T>::size() const [with T = signed char]' at ./core/templates/vector.h:99:57,
    inlined from 'Error GLTFDocument::_encode_buffer_view(Ref<GLTFState>, const double*, int, GLTFAccessor::GLTFAccessorType, GLTFAccessor::GLTFComponentType, bool, int, bool, GLTFBufferViewIndex&, bool)' at modules/gltf/gltf_document.cpp:1204:33:
./core/templates/cowdata.h:197:33: error: array subscript -1 is outside array bounds of 'signed char []' [-Werror=array-bounds=]
  197 |                         return *size;
      |                                 ^~~~
cc1plus: all warnings being treated as errors

It seems that the compiler is trying to inline this function, and giving a warning that it is out of bounds, even though it is not out of bounds. A simple fix is to store the buffer size in a variable. This also reduces code duplication. While I was at it, I also made old_size const.

This only occurs on arm64 Linux with GCC. It doesn't occur with scons use_llvm=yes on the same system, and it doesn't occur on any other platforms as far as I can see. This problem also only occurs with Vector<int8_t>, it works fine for all the other integer sizes, however I made the same change for all of them for consistency, plus just in case some other platform out there is incorrectly inlining and warning for a different integer size.

I'm very confident this doesn't change behavior, but I tested exporting and importing a .glb file and it still works correctly.

[akien-mga]

Thanks!