Use getenv to detect Nsight.

End can be equal to start in CalculateVertexArraysSize. This is quite
common when the vertex size is zero.

- Fixes renderdoc with OpenGL renderer.

The texture cache already takes care of setting a render target to dirty 
when invalidated.

This PR aims to reduce the memory usage in the CPU page table by moving
GPU specific parameters into a child class. This saves 1Gb of Memory for
most games.

Use INSERT_UNION_PADDING_WORDS instead of alignas to ensure a size
requirement.