Flaky test: headless_load_and_save_roundtrip

[mxgrey]

One of our unit tests is known to be flaky in GitHub's CI. It seems to be related to GPU drivers, even though the test is supposed to operate in headless mode. Most of the time it works as intended, but once in a while we need to trigger a rerun.

[mxgrey]

This may have been a false alarm. The most recent test failures were legitimate, not flaky.

[mxgrey]

Reopening because the race condition did show up again with this error message:

thread 'Async Compute Task Pool (0)' panicked at /home/runner/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/wgpu-hal-24.0.4/src/gles/device.rs:1445:39:
Could not lock adapter context. This is most-likely a deadlock.