Implement better index caching in Latte

[Exzap]

What this PR does:

• Implement a 8-entry LRU cache for indices, instead of only remembering the indices from the previous draw
• Reworked the indexData_* renderer interface a bit and added indexData_releaseIndexMemory to let the renderer know when cached index data won't be referenced anymore
• Implement this new interface on Vulkan. OpenGL uses a custom index cache so it's not affected

While working on this I realized that it's probably possible to share the buffer and heap allocator part of the logic between all three backends, so that the individual renderers would only have to implement buffer creation and uploads. But let that be a problem for the future.