Boost performance of groups with blend_mode != PassThrough

[Variable-ind]

NOTE: This is a performance PR so this would need a little testing, also it should be merged after Pixelorama 1.1.2 gets released (just a preference, not a requirement)

Attempt at solving #1250. Boosts the performance of GroupLayers to roughly twice (47% to be precise) it's current state by introducing cache.

• In consecutive generations, when the shader is same, calling RenderingServer.material_set_shader again and again (in generate_image()) takes up unnecessary computation time (Which makes up most of the 47% time i talked about -- per generation). There is now a way to reuse the material in last generation for the next one.

• when a GroupLayer gets blended, if it has another GroupLayer as child, even if the contents of child GroupLayer didn't change, it ends up being re-blended regardless. To fix this there is now a cache which save time if the contents of a GroupLayer didn't change.

As it uses cache this takes up a little more memory but it's not noticeable.

[OverloadedOrama]

I tested it a bit and it works well. I think it should be fine to get merged for v1.1.2, it's a very welcome performance boost and it doesn't seem to be breaking anything. Thank you very much!