Ask AI to generate tests for as many scenarios of potential race conditions for the whole framework as it can manage

[teremterem]

• Make sure to hit hard on intermixing of sync and async promising functions
• Ask it to first generate textual descriptions of scenarios (not all of them might be easy to generate tests for right away)
• Strip away all the thread-safety for the first pass to capture race conditions that are already safe-guarded against
• Bring back the existing thread-safety and make the AI to try harder to find ones that are still not safe-guarded against
• Ask to do third and fourth passes, just for a good measure (any more that are also safe-guarded against already plus any more that are still not safe-guarded against)