openssl deterministic test fails

[LCBH]

• Implementing PartialEq and Debug for TraceContext so that we can compare the two TraceContext obtained at different times from executing the same trace.
• Using that to test PUT's determinism after executing client_attacker_full.
• Currently fails for OpenSSL 111j. See the Issue OpenSSL is not made fully derministic #294 for this new failing uni test openssl::deterministic::tests::test_openssl_no_randomness_full.

[LCBH]

Commit #aa340 fixed the issue:

New determinism architecture:

• distinguish re-set RAND from reseed
• reset RAND once for all for all factories with determinism_set_reseed_all_factories
• reseed all factories before executing a trace with determinism_reseed_all_factories
  --> New integration test for determinism. Passes for OpenSSL!

The test test_attacker_full_det_recreate now passes for openssl111j.

[LCBH]

The test is actually not really reproducible. b1cdbbb is doing the same on multiple traces and it fails at some point :(
Sometimes at step 0, sometimes at step 20, 15, etc...
Why is so?

[LCBH]

Conclusion: this test fails because of session tickets sent after the final Server FInishes. Those contain time-related information, which thus differ across executions.

In BoringSSL, the flag BORINGSSL_UNSAFE_DETERMINISTIC_MODE reseed the PRNG but also make BoringSSL use a hard-coded time instead of the actual time. We have not found the equivalent for OpenSSL.
A solution would be to use libfaketime but this will also impact tlspuffin's timestamps.
One could also hijackossl_time_now but that's not really robust in the long term.

[LCBH]

Will close without merging, the useful more complete deterministic uni test is also in #290 that will me merged.