Don't guess at STHs

[rlb@…]

A call to get-sth-consistency should either return a consistency proof between
the two indicated tree heads or it should fail. The server should not change
the request to something it like better. If the front-end the client is talking
to is behind, then the client needs to wait until it catches up; an intermediate
STH is not useful.

Likewise, if a get-all-by-hash query can't provide the requested proof, it
should fail.

[eranm@…]

See discussion on issue 168 and Andrew Ayer's objection on the list:
​https://www.ietf.org/mail-archive/web/trans/current/msg02848.html

Given the justification for keeping it, suggest closing as wontfix.

[eranm@…]

I'll note this came up today again in an out-of-band discussion with colleagues: Having the option of getting the new STH by not specifying the 2nd tree size is an efficient way to catch up with the log without calling get-sth first - a single call is enough to get a new STH and a consistency proof to it from the STH the client currently knows about.