In-memory tables can't verify the uniqueness of a virtual generated column

[nicktobey]

Secondary indexes on in-memory tables are currently not very efficient: they're stored as a sorted list of rows, and don't get re-sorted until the end of a batch table edit operation.

This makes detecting duplicate values in a unique virtual column slow and nontrivial.

In order to do this, we would need to either:

• Scan every row of the table for each insert operation, computing the virtual column for each row.
• Use a data structure for indexes that can be updated in place

I added disabled tests for this in #2641