Small perf improvement to limited incremental sync

[erikjohnston]

The idea here is that we often end up fetching enough events from the DB to fill the gap, but we still end up with a gap due to the timeline limit. However, we can still apply the same optimisation as before to the list of fetched events, rather than just the truncated list we return.

[erikjohnston]

cc @richvdh, as this is adding another optimisation on top of the sync state delta work you/we did

[erikjohnston]

Thinking about it, I'm slightly worried that we don't have tests for this case. I'll need to have a think about how we go about doing that :thinking: