RPRBLND-1967: Improve "Sync process" for instances

[DagerD]

PURPOSE

Speed up sync process for final rendering

EFFECT OF CHANGE

Sync process for final rendering became much faster

TECHNICAL STEPS

• Implemented multi-threading for instanced objects
• Implemented USD instancing technique
• Added const CHUNK_COUNT to define how many object per thread will be processed (I figured out that 500 is almost ideal for my PC, I think we need to investigate which is better for other PCs)
• Added is_rand argument to function get_temp_file. We need it because USD doesn't update material if name of mtlx file isn't changed
• Modified blender_data node to fit new sync logic

NOTES FOR REVIEWERS

• https://github.com/GPUOpen-LibrariesAndSDKs/RadeonProRenderUSD/pull/512 is strictly needed for working with instancing.
• Here is a delay after syncing and before rendering which doesn't concern to sync process. The reason is HdRPR creates material for every instance (the same material).

[bnagirniak]

Waiting for https://github.com/GPUOpen-LibrariesAndSDKs/RadeonProRenderUSD/pull/512 to be merged

[bnagirniak]

@DagerD please fix conflicts. Waiting for #167 to be merged.

[bnagirniak]

167 is merged, now we can merge this PR