tablet_incremental_selector should be configured only with ranges within this replica

[Deexie]

Currently tablet_incremental_selector returns empty sstables vector if a range is not on this replica. It should not need to check the tablet's existence.

The responsibility for generating ranges fully contained within one tablet and skipping the empty one should rather be put on caller.

See discussion #17664 (comment) for the details.

[bhalevy]

What's the severity of this issue?
That it cause any problem other than inefficiency?
and if so, how bad is it?

[raphaelsc]

What's the severity of this issue? That it cause any problem other than inefficiency? and if so, how bad is it?

I think the most important thing is to unlock the ability to detect that a tablet is not present, during range scans, so we fail rather than returning wrong empty data.