MitM detection

[martinthomson]

Secondary certs depends on there not being a MitM. And sadly, they still exist in some places.

Now, as an advisory extra, the failure mode isn't fatal to the connection, but it could mean a lot of wasted effort. So we could pack the setting with 32 bits extracted from the exporter as confirmation that the mechanism works, or design a new mechanism to reduce the chance that we waste effort on creating authenticators that we can't use.

Either way, it's probably worth mentioning this possibility.

[MikeBishop]

Thirty-two bits from the exporter seems like a reasonable mitigation. That's fairly low cost.

[martinthomson]

The only wrinkle is that they might be all zero. If settings parsers return 0 and the exporter returns an all zero 32-bit value, then there is a possibility of a false positive.

I think that we can either force a bit to one and lose a bit, or emphasize that endpoints have to check for the presence of the setting as well as its value.