Suppose we have a table and a query:

CREATE TABLE a (a INT PRIMARY KEY, b INT, c INT, FAMILY(a), FAMILY(b), FAMILY(c);
INSERT INTO a VALUES(1,2,3);

SELECT b FROM a WHERE a=1;

Currently, this query will cause a scan from /a/primary/1 -> /a/primary/2, even though ideally we'd only need to scan the key at /a/primary/1/b.

We have enough information to improve this scan, since we know that c is not a needed column, a is the primary index and therefore unique, and we have an equality constraint on a.

We should consider improving the algorithm in spanFromLogicalSpan to incorporate all of this information into its final output span.

cc @danhhz @andreimatei @petermattis based on our earlier conversation.