These tests failed due to an indefinite hang in buffer.condition in the following code:
\`\`\`rust
    let buffer = cx
        .add_model(|cx| Buffer::new(0, cx.model_id() as u64, text).with_language(language, cx));
    buffer.condition(cx, |buf, _| !buf.is_parsing()).await;
`\`\`
In both gpui1 and gpui2 \`.with_language\` spawns a task that notifies the context once it's done. The \`condition\` waits for notifications to be raised. The gist of the problem was that in gpui2, the spawned task was scheduled straight away, so we never really saw the notification with \`condition\`, causing us to wait indefinitely. This is probably a difference in test between schedulers in gpui1 and gpui2, but I kind of sidestepped the issue by spawning a condition before firing off a parsing task with \`set_language\`.