Fix and test token aware routing

[conorbros]

Pre-req:

• Cassandra flags #853

Fixes a bug in token aware routing and tests against the Scylla driver.

[rukai]

I removed the bug fix and confirmed that causes the new test case to fail.

I also added some debug logs to verify that we were indeed routing to different nodes each time:

2022-10-23T23:56:02.855725Z ERROR lib::cassandra_int_tests::routing: delete hit: 172.16.1.3
2022-10-23T23:56:03.065194Z ERROR lib::cassandra_int_tests::routing: delete hit: 172.16.1.4
2022-10-23T23:56:03.274862Z ERROR lib::cassandra_int_tests::routing: delete hit: 172.16.1.3
2022-10-23T23:56:03.483010Z ERROR lib::cassandra_int_tests::routing: delete hit: 172.16.1.2
2022-10-23T23:56:03.691256Z ERROR lib::cassandra_int_tests::routing: delete hit: 172.16.1.2

The new routing.rs test case takes 84s to run on my machine.
Thats quite a hit to our total test runtime (+168s) and to the runtime of our standard cluster test case test_cluster_single_rack_v4 (84s)

With only 800 prepared queries that is 100ms per query, that seemed pretty bad to me?
But even when I skip shotover and query cassandra directly it still takes 83s, so seems cassandra is just slow 💤

Can we lower the loop to something like 0..10, yes it opens us up to a likelier chance of a false negative but that is not as likely as it first sounds since that still leaves us with 10 * 4 * 2 = 80 asserts and that number will grow as we expand our test suite to run against more scenarios.
(would also be down for 0..20 if 10 is pushing it)
We would also pretty quickly catch such cases as its bound to fail the next time we run tests.

[rukai]

Good job on coming up with a way to test that our routing and then figuring out what was wrong with our implementation.
Using tracing to ensure our routing matches that of a known good implementation is really clever!

[benbromhead]

Nice one - clever!!