test_hs24.test_zero_ts fails. It seems this got missed in WT-8287 because by default none of the asynchronous deletes it runs make it into the checkpoint it takes, and consequently the code to account for FLCS deleted values reading back as 0 didn't get added. Apparently the test machines are lagged enough to change this so it fails.

Solution is to add the missing logic.

I'm also going to add sleep(3) before the checkpoint (like some of the other tests that do simultaneous updating and checkpointing) so that the deletion thread has a chance to get going before the checkpoint starts. This makes it fail on FLCS (without the fix) even on an unloaded machine, and should make it more likely to turn up any other anomalies that might happen.