Once upon a time, it made sense for the ecpg preprocessor to have its
own version number, because it used a manually-maintained grammar that
wasn't always in sync with the core grammar.  But those days are
thankfully long gone, leaving only a maintenance nuisance behind.
Let's use the PG v10 version numbering changeover as an excuse to get
rid of the ecpg version number and just have ecpg identify itself by
PG_VERSION.  From the user's standpoint, ecpg will go from "4.12" in
the 9.6 branch to "10" in the 10 branch, so there's no failure of
monotonicity.

Discussion: <1471332659.4410.67.camel@postgresql.org>