Unsupported empty clustering key in COMPACT table

[nyh]

An empty clustering key (e.g., an empty string) is allowed in normal tables, and we have several tests for that in test/cql-pytest/test_empty.py. However, it turns out that in Cassandra the support for an empty clustering key in a table WITH COMPACT STORAGE is more subtle:

1. If the COMPACT STORAGE table has a single-column clustering key "c", an empty string value for the clustering key is not allowed, and Cassandra gives the following InvalidRequest error: Invalid empty or null value for column c
2. If the COMPACT STORAGE table has a compound clustering key, say "c1, c2", an empty string value for one or even both of the columns is allowed.

Scylla gets both cases wrong:

In the first case (single-column clustering key c), Scylla gives a misleading error message: Missing PRIMARY KEY part c, which doesn't say what went wrong (since "c" was not missing in the write, it was just an empty string).

In the second case (compound clustering key c1,c2), Scylla gives this same error message Missing PRIMARY KEY part c1 when c1 is set to an empty string - instead of supporting this case as Cassandra does.

By the way, as far as I can tell, this bizarre situation isn't documented anywhere as far as I can tell, and I don't know its justification (maybe there is no real justification - we have exactly the same situation in normal tables for empty single-column partition keys which CQL forbids but allows empty partition-key components).

This issue is reproduced by our tests:

• test_empty.py::test_insert_empty_string_key_compact (the first case, just a wrong error message)
• test_empty.py::test_insert_empty_string_compound_clustering_key_compact (the second case, more serious)

And also the following tests translated from the Cassandra test suite, in cassandra_tests/validation/operations/compact_storage_test.py::

• testEmptyRestrictionValue
• testEmptyRestrictionValueWithMultipleClusteringColumns
• testEmptyRestrictionValueWithOrderBy
• testEmptyRestrictionValueWithMultipleClusteringColumnsAndOrderBy