Faster backend init, Agility SDK, other changes

[jkunstwald]

• Speed up backend init
  • Fastpath for "first" GPU preference
  • Add option to defer queue init
  • Add option to defer and parallelize command list init
  • Other improvements

PSO compilations and resource creation can start immediately after "core init", significant time can be saved on the critical path by deferring the new subsystem inits.

• Add option to enable D3D12 Agility SDK

Allows usage of "future" / arbitrary D3D12 versions independent from Windows version, can ship own D3D12Core.dll

Minor

• Add cmd::set_global_profile_scope
  • Use as the very first command to override the per-cmdlist profile scope
• Use phi::buffer_address in commands instead of seperate buffer/offset fields
• Vulkan improvements
  • Simplify Vulkan layer/extension startup
  • Improve Vulkan scratch alloc mechanism
• Improve buffer size display in debug names ("14.3 MiB" instead of raw byte amount)
• Fix D3D12 shader table root constants (raytracing)
• Add minor missing DLL exports
• Add raytracing PSO debug names
• Fix MSVC DLL copies (CMake)
• Add option to supply own scratch buffers for RT AS builds
• Add Optick profile scope nesting (D3D12)

[Philip-Trettner]

eventually I would like add things like these to cc::format (not now though)