Highlight refs demonstrates a problem with some of the tagger cleanup that recently went in. Part of highlight refs is highly synchronous. Namely that we remove the tags on movement or edits immediately so that we don't show "stale" tags. This is problematic because the previous pr made it much more efficient to compute tags. So we could end up with clearing the tags and tehn very efficiently showing them again.

The fix here is in two parts. The first is that we make the delay for highlight refs closer to the perceived delay that we used to incur in the older less efficient system once we added up all the delays that used to have.

The second is that because we are clearign out the old values, we don't actually want the mode where we take the results we have and introduce them very soon after they're available. Instead, we explicitly attempt to cancel any computation work and only compute the results against the latest when new events come in. We initially had a batchign work queue for this. however, the batch was only ever up to two items (just to say if we were on the initial item on a subsequent item). because of that, it was just easier to go to simple task queuing to represent this concept of stopping the current set of work and just restarting the new computation

Manual testing in practice indicated that this feels good and non-distracting. With resutls still coming up in a reasonable amount of time, without it feeling aggressive.