[syncschema] don't arbitrarily remove constraints from the in-memory schema
For some constraint types, it may make sense to have several constraints
of that type simultaneously. In such a case, removing one of them from
the in-memory schema is wrong.
Closes #7446353
[server/test] don't leak timeout from one test class to the next
Reset SQL_CONNECT_HOOKS after we're done with this test class so the
next one doesn't inherit a timeout it doesn't want.