Optimise GL fence checking by querying less

[Dinnerbone]

Connections
This fixes #5179

Description
By keeping better track of which GL sync object last signalled, we can avoid repeatedly querying the status of them over and over. Additionally, bail early when encountering an unsignalled object as we know the remaining objects can't be signalled.

This has a noticable impact on performance when there are many Queue::submit()s per frame. For an extreme and contrived case, testing with the code below shows a reduction in frame time from 2 minutes to 2 seconds for me (with firefox, which seems to have a high cost of each sync check due to IPC)

(It's likely we can apply the same optimisation to Vulkan's FencePool too, but, I doubt the cost is nearly as high there as WebGL.)

Testing
Passes tests, and manually tested the impact by adding the following code to hello_triangle:

for _ in 0..1000 {
    let encoder = device.create_command_encoder(&Default::default());
    queue.submit(Some(encoder.finish()));
}

Checklist

• Run cargo fmt.
• Run taplo format.
• Run cargo clippy. If applicable, add:
  • --target wasm32-unknown-unknown
  • --target wasm32-unknown-emscripten
• Run cargo xtask test to run tests.
• Add change to CHANGELOG.md. See simple instructions inside file.

[cwfitzgerald]

Very nice improvement