Fix usage of index offsets in RenderingDevice

[pkdawson]

Fixes #86838

Removed duplicate usage of index offset. I tested Direct3D on Windows 11, it looks like it works the same way as Vulkan in this respect, so it also required a fix to render correctly.

[clayjohn]

The change makes sense to me. But I would like @RandomShaper to confirm that its fine to merge

[RandomShaper]

I acknowledge the issue boils down to the offset being applied twice. However, I'd like to state a different take on the solution.

To me, the drivers are doing the right thing. I mean, at their level of abstraction, they have no knowledge about what the offset is for, so they must just apply it. The problem lies at the RenderingDevice level, because that's the one doing the double-application. It binds with offset and then asks to render with the offset again. My suggested fix would be then to stop RenderingDevice from re-applying the offset at draw time.

This (untested) patch is what I'd do:
p.txt

[pkdawson]

As described previously in #68293, RenderingDevice::index_array_create takes the offset as a number of indices, but the drivers want the offset in bytes, so that needed an adjustment. Confirmed that this solution works too.

[clayjohn]

Looks great! Sorry for breaking this twice now 😬

[akien-mga]

This has conflicts following the merge of #84976, could you rebase (and I would suggest re-testing, as #84976 changed things significantly).

[pkdawson]

Rebased. Confirmed that this works with my MRP and my more complex test case of Dear ImGui meshes (https://github.com/pkdawson/imgui-godot), with no other obvious bugs.

[akien-mga]

Thanks!