I picked this up and worked on it somewhat recently, but stalled out due to the complexity. The complexity is mostly because it's hard to resist solving the juicy edge cases and autofixing everything, so next time I pick this up I'll just drop the complex parts, see if the basic cases that are working are worth having as a rule, and clean up the PR so it can be merged.

picked this one up to get it done and dusted. I tried to do simple solutions to avoid the complexity trap I previously fell into, but that only led to further problems and in the end I essentially needed to implement the reverse of the logic in 910/911/913.
The main simplification is avoiding any special handling with the ARTIFICIAL_STATEMENT in loops, but I'm not even sure that should be done as having (potentially "redundant") checkpoints in non-infinite loops can be a good idea.

I also replaced a lot of await foo() with await trio.lowlevel.checkpoint() in other ASYNC91X files in order to test them with ASYNC914 as well. This turned up several issues that I fixed, as well as highlight a case where the autofixer now needs two passes. If there's only a conditional checkpoint it won't be marked by ASYNC914 at first pass, but 910/911/913 will insert a conditional checkpoint, at which point the conditional checkpoint is now redundant. I did not bother fixing this.

I also ended up adding asyncio autofixing support to 910/911/913 by having them insert await asyncio.sleep(0), this ended up being very easy and made the testing easier.