Consider changing TCK tests so they do not interfere with each other or rely on ordering

[tjquinno]

Some of the existing tests use method ordering with the test class; it would be good to avoid the need for explicit ordering.

Also, some of the added tests for 5.0 reuse metric names across tests. If those tests are run independently, they work correctly. But if they are run in the same JVM without clearing out the registry(ies) between the executions of test classes, some of the tests fail or trigger failures in the implementation's validation checks.

Example failure: a test registers two counters and expects the total number of registered counters to be two. An alternative approach would be to find the count at the beginning of the test, then register the two counters, then make sure the new count is two more than the original count.

Example validation failure: A metric with a given name is registered with one set of tag names in one test and a different set of tag names in another test. The second attempt to register will fail (unless the registry has been cleared in the meantime) because of the inconsistent tag name set.