[3.x] Fix glGet overflows by using 64 bit versions

[lawnjelly]

Certain glGet operations require 64 bit versions according to the GLES spec. The previous code was susceptible to overflow bugs, especially running under ANGLE.

May fix #75913

Notes

• Supersedes Defend against invalid GL_MAX_UNIFORM_BLOCK_SIZE. #82921
• Also fixes a theoretical bug in gl_wrapper by placing an upper limit on texture image units.
• Is basically a slightly more in depth version of Fix integer value for GL_MAX_UNIFORM_BLOCK_SIZE overflowing #80909 .

[lawnjelly]

Let me know if there are any more than get spotted that should be 64 bit from the spec, I've tried to catch the ones from #80909 .

For the shaders, instead of reading max_image_units directly I've exposed the maximum desired so the calculations will be the same. We could alternatively e.g. make max_image_units static, or make storage readable from the shader file, or copy this value in the shader setup(), but it looks like more code for something that only needs a simple solution.

UBO Size

The UBO size seems to be the primary bug, and I have expanded some comments to discuss this. The problems seem to occur under ANGLE.

• This needs to be retrieved as 64 bit.
• Angle may report INT32_MAX + 1, a very high value suggesting that system memory is being used, but the hardware supported size may be lower, and we have no way of querying what it will be (afaik).

There is thus no ideal way to deal with this situation, and we have to compromise. We could e.g.:

1. Use the large value reported. This should not crash, but may end up being slower.
2. Detect the situation and revert to a value we know will work in hardware (the minimum), but may not be as fast as using the actual hardware max (which we don't know).

I've gone for (2) here, which still allowing for using larger sizes up to a megabyte, which we seem unlikely to use over in Godot 3.x.

[akien-mga]

Thanks!