Batch size >1000 does not pay off

We did a small experiment to see how a full scan of the Redis keys on
gitlab.com speeds up as we increase the batch size. The values on the
right are time in seconds for a full scan (no delete operations).

```
count: 10);      284.500529021
count: 100);      86.21216934
count: 1_000);    60.931676195
count: 10_000);   60.96355610
count: 100_000);  62.378172541
```

It looks like 1,000 is a good number.

See merge request !2970