perf: Do not use pointer map for tables with a unique constraint

[Centril]

No description provided.

[gefjon]

I would perhaps reframe this as making it so that every table has a single primary key index, which by default is on the tuple of all its columns, but can be narrowed to a single row or a smaller tuple. But that may be too large of a rearchitecting to make in our timeframe.

In the simpler ("hacky") version, Table::delete, aka delete_by_row_value, which devolves to Table::find_same_row, will need to contain additional logic to locate a unique index and use it to scan for the sought row. I am not concerned about the performance of this, but it may require significant fiddling just to make it work.

[Centril]

I would perhaps reframe this as making it so that every table has a single primary key index, which by default is on the tuple of all its columns, but can be narrowed to a single row or a smaller tuple. But that may be too large of a rearchitecting to make in our timeframe.

That sounds like a nice way of looking at it, but as you say, doing the hack might be more within our time-constraints, for now. I think I'll try the hack first and then we can make it pretty later on.