Prefetcher use in chunk apply could be improved

[nagisa]

Recently we've made some improvements to the prefetcher such that it prefetches data that's necessary for chunk finalization to progress a little bit faster. That also results in less deviation in chunk apply times, and thus, probably, also fewer missed chunks overall.

Anyway, #10935 points out multiple points of improvements we could pursue:

• prefetch receipt data earlier (especially for delayed receipts -- earlier than right before receipt execution begins;) (Move delayed receipt (and possibly other receipt) prefetching to the beginning of apply #11002)
• have multiple levels of prefetch priority, so that e.g. prefetching for finalize doesn't interfere with prefetch requests for data that is going to be needed sooner;
• figure out how to prevent overfetching by as much (perhaps adding a way/guards to remove individual keys rather than clearing the full prefetch queue?)