sstable: include count of iteration steps that saw the same prefix

In https://cockroachlabs.slack.com/archives/C01RV1QS88N/p1755224653267019, we were able to use the separated count stats to infer that we were seeing two version for each `roachpb.Key`.
```
n1 scan stats: stepped 85.612k times (85.623k internal); seeked 2 times (2 internal); block-bytes: (total 1.4 MiB, cached 1.4 MiB, duration 0s); points: (count 85.625k, key-bytes 1.6 MiB, value-bytes 146 KiB, tombstoned: 0) ranges: (count 0), (contained-points 0, skipped-points 0) evaluated requests: 0 gets, 1 scans, 0 reverse scans separated: (count: 42.165k, bytes: 214 KiB, bytes-fetched: 0 B)
```
With value separation using blob files, value blocks are not the only place we will be placing older versions, so such inferencing would no longer be possible.

It would be better to have an explicit count of duplicated prefixes, encountered using iteration, and we can use the `prefixChanged` bitmap to compute this.

Jira issue: PEBBLE-1146