Swapchain performance/correctness

[attackgoat]

Modifies the Display struct to own a Swapchain, instead of operating in conjunction with it. Fixes the incorrect "image ready" fence which was waited upon before submitting new frames.

The previous behavior was unsound and only worked on certain hardware.

The new pattern greatly simplifies swapchain handling:

• A list of command buffers available to execute frame submissions is kept in Display
• The list of command buffers may have 1 or more items, up to and beyond the number of swapchain images

Process:

• vkAcquireNextImageKHR signals an "acquire" semaphore when the image is available
• Rendering commands wait for "acquire" and signal a "rendered" semaphore
• vkQueuePresentKHR waits for "rendered" before presenting the image

Additionally, recreating the swapchain no longer uses vkDeviceWaitIdle. This improves window resizing performance.

[attackgoat]

Tested platforms:

• Windows Nvidia
• Windows Intel Iris Xe
• X11 Nvidia
• Wayland Nvidia
• Apple M3

Notes:

• Found unrelated issues in multithread.rs and egui.rs which cause segfaults on Windows and Wayland
• Resizing the window on Apple causes a segfault - this will be solved later

Platform support:

• Windows and X11: All bugs should be solved ASAP
• Apple: Some bugs will exist and be solved as time permits
• Wayland: Best effort, but it doesn't run Chrome properly so
• Android: Should work, I welcome someone to help verify this