DX12 Swapchain buffers' resource states could be wrong for trimming case.

[locke-lunarg]

Now, during capture, only GFXRECON_CAPTURE_DRAW_CALLS case(https://github.com/LunarG/gfxreconstruct/pull/1680) writes swapchain buffers. The other trimmings, that frames and queue submits, don't write swapchain buffers. It also doesn't write their resource states (MetaDataType::kInitSubresourceCommand). In this case, during replay, swapchain buffers aren't initialized to the correct resource states. It could happen wrong resource state errors.

We could add a new MetaDataType, like kInitSwapchainResourceStatesCommand, to fix it. Or we could wait until the new architecture that replay could access capture to fix it.