Extend array rapidcheck templates to support nullable and variable-length fields

[rroelke]

Resolves CORE-23.

#5417 added a suite of tests which create an array with random data, and then read back the data and compare against the input.

Those tests were limited to single-valued non-nullable attributes. This pull request extends their capabilities to support nullable attributes and variable-length attributes (and nullable variable-length attributes).

Previously the code all used std::vector<T> to represent the single-valued attribute data. Now to represent the set of buffers needed to write or read a user type, we use struct query_buffers<T> where T some logical representation of the user type:

• some fundamental type T for a non-nullable single-valued cell
• std::vector<T> for a fundamental type T for a non-nullable variable-length cell
• std::optional<T> for a fundamental type T for a nullable single-valued cell
• std::optional<std::vector<T>> for a fundamental type T for a nullable variable-length cell

The specializations of query_buffers contain the columnar buffers for each of these.

The run_execute function which drives the queries does multi-part reads into a large query_buffers which it then compares against the expected result. To handle the variable-length data it is necessary to update the uint64_t outcursor which previously just held the number of cells already read to a tuple which contains the variable-length position as well as the cell position.

Because this is done with static dispatch there's a lot of std::apply which basically means "turn this tuple into a variadic function argument list".

Testing

One of the existing tests is updated to use a non-nullable variable-length attribute and one is updated to use a nullable variable-length attribute.

---

TYPE: NO_HISTORY
DESC: Enable nullable/variable data in array rapidcheck tests

[ypatia]

Thank you for the detailed description, that helped a lot understanding what's in here.