Primitive Keys: Remove WITH(KEY='z')

[big-andy-coates]

With #3536 complete we can deprecate the use of WITH(KEY='x') in all statements, as the user can now use whatever name they like for their key column.

Internally, we can still choose to maintain a track of fields in the value that are copies of the key. However, this becomes an internal optimization. However, I think the ROI on this is very low. We already know there are a lot of bugs in the key field handling code and it adds a LOT of complexity to KSQL. Plus, to support this properly we'd need to extend the key field functionality to be able to track multiple copies of the key in the value, which is more work and more complexity and more space for bugs.

The way I see it is that the WITH(KEY) syntax was introduced to allow users to use a nicely named column in their SQL statements, e.g.

CREATE TABLE FOO (ID INT, NAME STRING) WITH(KEY='ID', ...);
SELECT COUNT() FROM FOO GROUP BY ID;

In the above select the user can use ID in the GROUP BY, which is more descriptive that GROUP BY ROWKEY, (which is equivalent).

However, with the ability to name the key column you can now do:

CREATE TABLE FOO (ID INT KEY, NAME STRING) WITH(...);
SELECT COUNT() FROM FOO GROUP BY ID;

Which has the same effect, without the need to duplicate the key into the value.

[big-andy-coates]

As part of this work we should update partition-data.rst