Should `beforeinput` be fired for composition? If so what should the target ranges be?

Currently, Chrome does not fire beforeinput for composed input for EditContext (but does for other editors) — I wonder if that is intentional or a bug… The spec as it is currently written seems to indicate that it should be.

For composed text input, we can try to infer where the composed text is, but that could easily be buggy, since there may be special formatting or it may be split across multiple pages in the case of a document editor, etc. So maybe the best solution is to just have it be an empty array, like we do for <canvas>, assuming we want to have the beforeinput at all (I think probably we do, to be consistent with contenteditable…).

Start by comparing Chrome's EditContext behavior with the current specification wording for beforeinput during composed text input. Examine the existing canvas behavior for target ranges and determine whether composition should fire beforeinput and use an empty array or another range representation; done requires a resolved, spec-consistent decision.