Port to gfx-rs HAL

[kvark]

Current three-rs is based on the published gfx-0.17, which is developed in pre-ll branch of gfx-rs. The new gfx-rs core is called HAL, and it's first version is about to be published, so we should discuss the prospects on moving into it.

Pros:

• super helpful dog-fooding for HAL
• potentially faster
  • requires more architectural changes to take advantage of
  • counter point: we are fast enough for the target use cases
• easier to extend with user data because of descriptor sets
  • counter point: user needs to understand parts of HAL

Cons:

• HAL is not as stable as the current gfx, both in terms of API and run-time bugs/features
• considerable amount of work to port
  • largely depends on helper layers (like gfx_mem) that we need to use, and they are even less stable at the moment

[kvark]

My current position: super fine control of latency and CPU overhead is not required for Three-rs. We should focus instead on features and ergonomics for now. Then re-consider later when HAL is more polished and accessible.

[kvark]

Closing in favor of #222