fix(s2n_session_ticket_test): correct clock mocking

[jmayclin]

Resolved issues:

Perhaps addresses #4600

Description of changes:

We have observed that the session_ticket_test is flaky. I have a hypothesis that the flakiness occurs under this scenario

1. s2n_config_add_ticket_crypto_key is called with intro time of 0
2. wall_clock is called to set the intro time of the key, let say 1_000 ns.
3. during the handshake, s2n_get_ticket_encrypt_decrypt_key is called to get an encryption key
4. wall_clock returns the same time, or a smaller time <= 1_000 ns which is possible because wall clocks can move backwards.
5. therefore no encryption key is found
6. and no session ticket is sent.

We fix this with two changes
First, we mock the time before adding the ticket key, ensuring that we know exactly what the ticket intro time is.
Secondly, we allow keys to be used at their intro time, rather than enforcing keys be past their intro time.

Call-outs:

It's incredibly difficult to test whether this is causing the actual flakiness, but by mocking the wall clock I can confirm that the above sequence of events is 100% possible.

Testing:

All CI should pass. By mocking the wall clock at the start of the test, I now have test coverage over the <= condition that I added.

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

[lrstewart]

Did you try running the test in a loop for a while? Were you able to repro the failure at all?

[jmayclin]

I did try running the test in a loop, but was unable to reproduce the failure locally. 😢

So this definitely fixes a theoretical problem, and maybe a real one, so I'm content to move ahead with it.

If we were fine with artifical clock skew, I could add fake noise to the wall clock measurement, and then my changes would successfully fix it.