Mostly fix a failing Test_EmptyColumnValues test, but also add a new typed failing check in same test

[sl-at-ibm]

This PR serves as a suggestion to work around a regression silently introduced on main.
(silent because the test was disabled on HCD and did not get caught).

So, a (sound, pre-GA) solution is offered for the current failure, but also a typed-read-path new problem is highlighted, which I think should be addressed pre-GA. In more detail:

• enable Test_EmptyColumnValues test for HCD. The test behaves identically on Astra/HCD. (note for the future: are there any other tests that are Astra-only for no reason?)
• solve the untyped check on Genres (I think it comes from the fact that untyped now returns JsonElements. A broader typecast resolved the failing check and I think it's ok from a untyped perspective)
• PRE-GA IMPORTANT Added a section of the test that currently fails. It should be addressed pre-ga. My expectation is that reading a table with a "nullempty" set, but using a typed approach with a non-nullable HashSet for the field, should populate the row with an empty HashSet and not a null. Yet, the test shows one finds a Null there <--- would you agree this should be changed?