schemachanger: better concurrency testing

[postamar]

When interpreting DDL statements, the declarative schema changer checks that there are no concurrent schema changes on the affected descriptors. This effectively acquires an exclusive lock on those descriptors.

We've had bugs like #106487 for which it has been difficult to validate the fix other than by running a workload, which is unfortunate. We should instead have an integration test in the form of a go test, in which a bunch of various schema changes fire up concurrently on the same limited set of descriptors.

For example:

1. for each table, one thread in which we sequentially rename a table by cycling through a bunch of names,
2. for each table, one thread in which we sequentially create and drop secondary indexes,
3. for each schema and for each database, cycle through a bunch of names
4. etc.

The fully qualified names of the tables can be retrieved in each transaction and fmt.Sprint-ed into the DDL statement by looking up the descriptors by ID because the IDs don't change.

This is just a general idea, we haven't implemented RENAME TO statements in the declarative schema changer as of yet.

Jira issue: CRDB-29951