Fix CI test flakes

[Fishbowler]

Herein lie attempts to fix the recent test flakes in CI. I believe them all to be reasonable changes or reasonable compromises.

[guusdk]

I don't really like the blunt retries - it hides a problem to the point where having the test may not reliably prove much any more.

I was considering to simply disable these tests, until we can get them more stable.

Happy to test these changes to see if they make a difference, but meh - I don' have high hopes.

[guusdk]

I don't really like the blunt retries - it hides a problem to the point where having the test may not reliably prove much any more.

I was considering to simply disable these tests, until we can get them more stable.

Happy to test these changes to see if they make a difference, but meh - I don' have high hopes.

[Fishbowler]

I don't really like the blunt retries - it hides a problem to the point where having the test may not reliably prove much any more.

Me neither, but in this case, it feels like a necessary evil, and didn't feel altogether all that evil....

Retries are added to 2 places:

• on the SSL socket layer, where the retry is dealing with a test setup race condition. We've already got the same line in RemoteReceivingServerDummy
• for a single retry on the outgoing test, where the expectation is testing combinations of connection logic for an outcome. If something is broken, it's likely always broken. Whilst it might fail for other reasons, it's unlikely to "accidentally pass". It's likely only hiding some test-only warm-up bug.