timeline: aggregated events aren't well ordered

Nobody has claimed this yet.

Aggregation events (edits, reactions) aren't well ordered in the timeline: we don't really know how to order them relative to each other. In the general case, it won't be a problem, but it might lead to discrepancies in a few cases:

- when we have a bundled edit, for instance, right now we'll always prefer it to a pending edit. This might be wrong if the pending edit is more up-to-date than the bundled edit (i.e. the related pending edit is *after* the bundled edit event, in the overall ordering).
- when we already have a pending edit and another comes in, we do some best-effort guessing if the new pending edit overrides the previous one or would be overridden by the previous one.

Ideally, the protocol would give us some perfect ordering over events, so we don't have to do any kind of guessing / rejiggering ourselves, e.g. with something akin to [MSC4033](https://github.com/matrix-org/matrix-spec-proposals/pull/4033). Otherwise, we may be able to use the `all_events` internal ordering (or the ordering provided by the [Event cache](https://github.com/matrix-org/matrix-rust-sdk/issues/3058), in some future).

No files or tests are named. Start by tracing timeline aggregation and the all_events ordering mentioned in the issue, then review MSC4033 and Event cache issue 3058. Done means the ordering of aggregated and pending events is deterministic and no longer relies on the described best-effort guesses.

Written by the indexing model from the issue text.