Add fold method to Batch and PreparedQueryIter

[newDINO]

Add their fold implementation and benchmarks.

For PreparedQueryIter, there is a −67% time decrease: from 25us to 8.1us.
For Batch, there is basically no improvement. But due to the uncertainty of compiler optimization, I think this may still be useful.

[newDINO]

I also think it is worth mentioning that for_each has better performance than for-loop in the documentation, especially when the there is no much computation per iteration, and it can achieve performance close to iterating over rust built-in slices.

bevy_ecs also has this kind of mentioning (https://docs.rs/bevy_ecs/latest/bevy_ecs/system/struct.Query.html#iteratorfor_each).

[Ralith]

Thanks! If you're currently using prepared queries, I recommend trying without them now that query caching is automatic.

[newDINO]

Thanks for the recommendation. Currently I'm not using PreparedQuery due to its !Sync limitations and that it's not very ergonomic.

On my machine, the automatically cached query only takes tens of nanoseconds, which is highly efficient.