Add async iteration to the cache API

Retreiving all the keys on a large cache can be very very slow, or in some browsers (as filed against Chrome https://bugs.chromium.org/p/chromium/issues/detail?id=688268) can just throw an exception.

This happened to us as our cache pruning algorithm uses the keys of the cache to determine which files are safe to delete and which must be kept. (It's perhaps an odd use-case, we're building an email client in the browser, and the attachment cache stores not only downloaded attachments which we can trivially re-download, but also uploaded attachments for which it is currently the primary source; we'll move the latter out so we can just call caches.delete(), but this seems likely to bite other people after a while in production).

Start by reviewing the Cache API's current key-retrieval behavior and the linked Chromium issue 688268. The issue names no repository files, tests, or implementation entry points, so first determine where the API is specified and how large-cache iteration is represented. Done means a concrete async-iteration change is defined and its behavior is validated for large caches.